read and write speech to text not working

Dictate your documents in Word

Dictation lets you use speech-to-text to author content in Microsoft 365 with a microphone and reliable internet connection. It's a quick and easy way to get your thoughts out, create drafts or outlines, and capture notes. 

Start speaking to see text appear on the screen.

How to use dictation

Tip:  You can also start dictation with the keyboard shortcut:  ⌥ (Option) + F1.

Learn more about using dictation in Word on the web and mobile

Dictate your documents in Word for the web

Dictate your documents in Word Mobile

What can I say?

In addition to dictating your content, you can speak commands to add punctuation, navigate around the page, and enter special characters.

You can see the commands in any supported language by going to  Available languages . These are the commands for English.

Punctuation

Navigation and selection, creating lists, adding comments, dictation commands, mathematics, emoji/faces, available languages.

Select from the list below to see commands available in each of the supported languages.

• Select your language

Arabic (Bahrain)

Arabic (Egypt)

Arabic (Saudi Arabia)

Croatian (Croatia)

Gujarati (India)

• Hebrew (Israel)
• Hungarian (Hungary)
• Irish (Ireland)

Marathi (India)

• Polish (Poland)
• Romanian (Romania)
• Russian (Russia)
• Slovenian (Slovenia)

Tamil (India)

Telugu (India)

• Thai (Thailand)
• Vietnamese (Vietnam)

More Information

Spoken languages supported.

By default, Dictation is set to your document language in Microsoft 365.

We are actively working to improve these languages and add more locales and languages.

Supported Languages

Chinese (China)

English (Australia)

English (Canada)

English (India)

English (United Kingdom)

English (United States)

French (Canada)

French (France)

German (Germany)

Italian (Italy)

Portuguese (Brazil)

Spanish (Spain)

Spanish (Mexico)

Preview languages *

Chinese (Traditional, Hong Kong)

Chinese (Taiwan)

Dutch (Netherlands)

English (New Zealand)

Norwegian (Bokmål)

Portuguese (Portugal)

Swedish (Sweden)

Turkish (Turkey)

* Preview Languages may have lower accuracy or limited punctuation support.

Dictation settings

Click on the gear icon to see the available settings.

Spoken Language:  View and change languages in the drop-down

Microphone: View and change your microphone

Auto Punctuation:  Toggle the checkmark on or off, if it's available for the language chosen

Profanity filter:  Mask potentially sensitive phrases with ***

Tips for using Dictation

Saying “ delete ” by itself removes the last word or punctuation before the cursor.

Saying “ delete that ” removes the last spoken utterance.

You can bold, italicize, underline, or strikethrough a word or phrase. An example would be dictating “review by tomorrow at 5PM”, then saying “ bold tomorrow ” which would leave you with "review by tomorrow at 5PM"

Try phrases like “ bold last word ” or “ underline last sentence .”

Saying “ add comment look at this tomorrow ” will insert a new comment with the text “Look at this tomorrow” inside it.

Saying “ add comment ” by itself will create a blank comment box you where you can type a comment.

To resume dictation, please use the keyboard shortcut ALT + `  or press the Mic icon in the floating dictation menu.

Markings may appear under words with alternates we may have misheard.

If the marked word is already correct, you can select  Ignore .

This service does not store your audio data or transcribed text.

Your speech utterances will be sent to Microsoft and used only to provide you with text results.

For more information about experiences that analyze your content, see Connected Experiences in Microsoft 365 .

Troubleshooting

Can't find the dictate button.

If you can't see the button to start dictation:

Make sure you're signed in with an active Microsoft 365 subscription

Dictate is not available in Office 2016 or 2019 for Windows without Microsoft 365

Make sure you have Windows 10 or above

Dictate button is grayed out

If you see the dictate button is grayed out

Make sure the note is not in a Read-Only state.

Microphone doesn't have access

If you see "We don’t have access to your microphone":

Make sure no other application or web page is using the microphone and try again

Refresh, click on Dictate, and give permission for the browser to access the microphone

Microphone isn't working

If you see "There is a problem with your microphone" or "We can’t detect your microphone":

Make sure the microphone is plugged in

Test the microphone to make sure it's working

Check the microphone settings in Control Panel

Also see How to set up and test microphones in Windows

On a Surface running Windows 10: Adjust microphone settings

Dictation can't hear you

If you see "Dictation can't hear you" or if nothing appears on the screen as you dictate:

Make sure your microphone is not muted

Adjust the input level of your microphone

Move to a quieter location

If using a built-in mic, consider trying again with a headset or external mic

Accuracy issues or missed words

If you see a lot of incorrect words being output or missed words:

Make sure you're on a fast and reliable internet connection

Avoid or eliminate background noise that may interfere with your voice

Try speaking more deliberately

Check to see if the microphone you are using needs to be upgraded

Need more help?

Want more options.

Explore subscription benefits, browse training courses, learn how to secure your device, and more.

Microsoft 365 subscription benefits

Microsoft 365 training

Microsoft security

Accessibility center

Communities help you ask and answer questions, give feedback, and hear from experts with rich knowledge.

Ask the Microsoft Community

Microsoft Tech Community

Windows Insiders

Microsoft 365 Insiders

Was this information helpful?

Thank you for your feedback.

How to use speech to text in Microsoft Word

Speech to text in Microsoft Word is a hidden gem that is powerful and easy to use. We show you how to do it in five quick and simple steps

Master the skill of speech to text in Microsoft Word and you'll be dictating documents with ease before you know it. Developed and refined over many years, Microsoft's speech recognition and voice typing technology is an efficient way to get your thoughts out, create drafts and make notes.

Just like the best speech to text apps that make life easier for us when we're using our phones, Microsoft's offering is ideal for those of us who spend a lot of time using Word and don't want to wear out our fingers or the keyboard with all that typing. While speech to text in Microsoft Word used to be prone to errors which you'd then have to go back and correct, the technology has come a long way in recent years and is now amongst the best text-to-speech software .

Regardless of whether you have the best computer or the best Windows laptop , speech to text in Microsoft Word is easy to access and a breeze to use. From connecting your microphone to inserting punctuation, you'll find everything you need to know right here in this guide. Let's take a look...

How to use speech to text in Microsoft Word: Preparation

The most important thing to check is whether you have a valid Microsoft 365 subscription, as voice typing is only available to paying customers. If you’re reading this article, it’s likely your business already has a Microsoft 365 enterprise subscription. If you don’t, however, find out more about Microsoft 365 for business via this link . 

The second thing you’ll need before you start voice typing is a stable internet connection. This is because Microsoft Word’s dictation software processes your speech on external servers. These huge servers and lighting-fast processors use vast amounts of speech data to transcribe your text. In fact, they make use of advanced neural networks and deep learning technology, which enables the software to learn about human speech and continuously improve its accuracy. 

These two technologies are the key reason why voice typing technology has improved so much in recent years, and why you should be happy that Microsoft dictation software requires an internet connection. 

Once you’ve got a valid Microsoft 365 subscription and an internet connection, you’re ready to go!

Are you a pro? Subscribe to our newsletter

Sign up to the TechRadar Pro newsletter to get all the top news, opinion, features and guidance your business needs to succeed!

Step 1: Open Microsoft Word

Simple but crucial. Open the Microsoft Word application on your device and create a new, blank document. We named our test document “How to use speech to text in Microsoft Word - Test” and saved it to the desktop so we could easily find it later.

Step 2: Click on the Dictate button

Once you’ve created a blank document, you’ll see a Dictate button and drop-down menu on the top right-hand corner of the Home menu. It has a microphone symbol above it. From here, open the drop-down menu and double-check that the language is set to English.

One of the best parts of Microsoft Word’s speech to text software is its support for multiple languages. At the time of writing, nine languages were supported, with several others listed as preview languages. Preview languages have lower accuracy and limited punctuation support.

Step 3: Allow Microsoft Word access to the Microphone

If you haven’t used Microsoft Word’s speech to text software before, you’ll need to grant the application access to your microphone. This can be done at the click of a button when prompted.

It’s worth considering using an external microphone for your dictation, particularly if you plan on regularly using voice to text software within your organization. While built-in microphones will suffice for most general purposes, an external microphone can improve accuracy due to higher quality components and optimized placement of the microphone itself.

Step 4: Begin voice typing

Now we get to the fun stuff. After completing all of the above steps, click once again on the dictate button. The blue symbol will change to white, and a red recording symbol will appear. This means Microsoft Word has begun listening for your voice. If you have your sound turned up, a chime will also indicate that transcription has started. 

Using voice typing is as simple as saying aloud the words you would like Microsoft to transcribe. It might seem a little strange at first, but you’ll soon develop a bit of flow, and everyone finds their strategies and style for getting the most out of the software. 

These four steps alone will allow you to begin transcribing your voice to text. However, if you want to elevate your speech to text software skills, our fifth step is for you.

Step 5: Incorporate punctuation commands

Microsoft Word’s speech to text software goes well beyond simply converting spoken words to text. With the introduction and improvement of artificial neural networks, Microsoft’s voice typing technology listens not only to single words but to the phrase as a whole. This has enabled the company to introduce an extensive list of voice commands that allow you to insert punctuation marks and other formatting effects while speaking. 

We can’t mention all of the punctuation commands here, but we’ll name some of the most useful. Saying the command “period” will insert a period, while the command “comma” will insert, unsurprisingly, a comma. The same rule applies for exclamation marks, colons, and quotations. If you’d like to finish a paragraph and leave a line break, you can say the command “new line.” 

These tools are easy to use. In our testing, the software was consistently accurate in discerning words versus punctuation commands.

Microsoft’s speech to text software is powerful. Having tested most of the major platforms, we can say that Microsoft offers arguably the best product when balancing cost versus performance. This is because the software is built directly into Microsoft 365, which many businesses already use. If this applies to your business, you can begin using Microsoft’s voice typing technology straight away, with no additional costs. 

We hope this article has taught you how to use speech to text software in Microsoft Word, and that you’ll now be able to apply these skills within your organization. 

Webflow announces acquisition of Intellimize - expanding beyond visual development to become an integrated Website Experience Platform

Square Online review 2024: Top ecommerce platform pros, cons, and features tested

I met two incredible teenage Swift coders who might care more about climate change than you

Most Popular

• 2 Google rolls out Pixel 8 and 7 update to finally fix their cellular network issues
• 3 Today's Wordle answer is the hardest this year, with an average score of 5.4, and 'Wordle 1037 X' is trending on Twitter – here's why it's so tough and what to do in future
• 4 Meta’s massive OS announcement is more exciting than a Meta Quest 4 reveal, and VR will never be the same again
• 5 Steve Jobs was wrong about the post-PC era and the next batch of iPads should embrace this
• 2 Missed out on the Fujifilm X100VI? New leak suggests the next best thing could land soon
• 3 Meta’s massive OS announcement is more exciting than a Meta Quest 4 reveal, and VR will never be the same again
• 4 Steve Jobs was wrong about the post-PC era and the next batch of iPads should embrace this
• 5 A full list of Samsung Galaxy Z Flip 6 and Z Fold 6 colors has leaked

Speech to Text - Voice Typing & Transcription

Take notes with your voice for free, or automatically transcribe audio & video recordings. secure, accurate & blazing fast..

~ Proudly serving millions of users since 2015 ~

I need to >

Dictate Notes

Start taking notes, on our online voice-enabled notepad right away, for free.

Transcribe Recordings

Automatically transcribe (and optionally translate) audios & videos - upload files from your device or link to an online resource (Drive, YouTube, TikTok or other). Export to text, docx, video subtitles and more.

Speechnotes is a reliable and secure web-based speech-to-text tool that enables you to quickly and accurately transcribe your audio and video recordings, as well as dictate your notes instead of typing, saving you time and effort. With features like voice commands for punctuation and formatting, automatic capitalization, and easy import/export options, Speechnotes provides an efficient and user-friendly dictation and transcription experience. Proudly serving millions of users since 2015, Speechnotes is the go-to tool for anyone who needs fast, accurate & private transcription. Our Portfolio of Complementary Speech-To-Text Tools Includes:

Voice typing - Chrome extension

Dictate instead of typing on any form & text-box across the web. Including on Gmail, and more.

Transcription API & webhooks

Speechnotes' API enables you to send us files via standard POST requests, and get the transcription results sent directly to your server.

Zapier integration

Combine the power of automatic transcriptions with Zapier's automatic processes. Serverless & codeless automation! Connect with your CRM, phone calls, Docs, email & more.

Android Speechnotes app

Speechnotes' notepad for Android, for notes taking on your mobile, battle tested with more than 5Million downloads. Rated 4.3+ ⭐

iOS TextHear app

TextHear for iOS, works great on iPhones, iPads & Macs. Designed specifically to help people with hearing impairment participate in conversations. Please note, this is a sister app - so it has its own pricing plan.

Audio & video converting tools

Tools developed for fast - batch conversions of audio files from one type to another and extracting audio only from videos for minimizing uploads.

Our Sister Apps for Text-To-Speech & Live Captioning

Complementary to Speechnotes

Reads out loud texts, files & web pages

Reads out loud texts, PDFs, e-books & websites for free

Speechlogger

Live Captioning & Translation

Live captions & translations for online meetings, webinars, and conferences.

Need Human Transcription? We Can Offer a 10% Discount Coupon

We do not provide human transcription services ourselves, but, we partnered with a UK company that does. Learn more on human transcription and the 10% discount .

Dictation Notepad

Start taking notes with your voice for free

Speech to Text online notepad. Professional, accurate & free speech recognizing text editor. Distraction-free, fast, easy to use web app for dictation & typing.

Speechnotes is a powerful speech-enabled online notepad, designed to empower your ideas by implementing a clean & efficient design, so you can focus on your thoughts. We strive to provide the best online dictation tool by engaging cutting-edge speech-recognition technology for the most accurate results technology can achieve today, together with incorporating built-in tools (automatic or manual) to increase users' efficiency, productivity and comfort. Works entirely online in your Chrome browser. No download, no install and even no registration needed, so you can start working right away.

Speechnotes is especially designed to provide you a distraction-free environment. Every note, starts with a new clear white paper, so to stimulate your mind with a clean fresh start. All other elements but the text itself are out of sight by fading out, so you can concentrate on the most important part - your own creativity. In addition to that, speaking instead of typing, enables you to think and speak it out fluently, uninterrupted, which again encourages creative, clear thinking. Fonts and colors all over the app were designed to be sharp and have excellent legibility characteristics.

Example use cases

• Voice typing
• Writing notes, thoughts
• Medical forms - dictate
• Transcribers (listen and dictate)

Transcription Service

Start transcribing

Fast turnaround - results within minutes. Includes timestamps, auto punctuation and subtitles at unbeatable price. Protects your privacy: no human in the loop, and (unlike many other vendors) we do NOT keep your audio. Pay per use, no recurring payments. Upload your files or transcribe directly from Google Drive, YouTube or any other online source. Simple. No download or install. Just send us the file and get the results in minutes.

• Transcribe interviews
• Captions for Youtubes & movies
• Auto-transcribe phone calls or voice messages
• Students - transcribe lectures
• Podcasters - enlarge your audience by turning your podcasts into textual content
• Text-index entire audio archives

Key Advantages

Speechnotes is powered by the leading most accurate speech recognition AI engines by Google & Microsoft. We always check - and make sure we still use the best. Accuracy in English is very good and can easily reach 95% accuracy for good quality dictation or recording.

Lightweight & fast

Both Speechnotes dictation & transcription are lightweight-online no install, work out of the box anywhere you are. Dictation works in real time. Transcription will get you results in a matter of minutes.

Super Private & Secure!

Super private - no human handles, sees or listens to your recordings! In addition, we take great measures to protect your privacy. For example, for transcribing your recordings - we pay Google's speech to text engines extra - just so they do not keep your audio for their own research purposes.

Health advantages

Typing may result in different types of Computer Related Repetitive Strain Injuries (RSI). Voice typing is one of the main recommended ways to minimize these risks, as it enables you to sit back comfortably, freeing your arms, hands, shoulders and back altogether.

Saves you time

Need to transcribe a recording? If it's an hour long, transcribing it yourself will take you about 6! hours of work. If you send it to a transcriber - you will get it back in days! Upload it to Speechnotes - it will take you less than a minute, and you will get the results in about 20 minutes to your email.

Saves you money

Speechnotes dictation notepad is completely free - with ads - or a small fee to get it ad-free. Speechnotes transcription is only $0.1/minute, which is X10 times cheaper than a human transcriber! We offer the best deal on the market - whether it's the free dictation notepad ot the pay-as-you-go transcription service.

Dictation - Free

• Online dictation notepad
• Voice typing Chrome extension

Dictation - Premium

• Premium online dictation notepad
• Premium voice typing Chrome extension
• Support from the development team

Transcription

$0.1 /minute.

• Pay as you go - no subscription
• Audio & video recordings
• Speaker diarization in English
• Generate captions .srt files
• REST API, webhooks & Zapier integration

Compare plans

Privacy policy.

We at Speechnotes, Speechlogger, TextHear, Speechkeys value your privacy, and that's why we do not store anything you say or type or in fact any other data about you - unless it is solely needed for the purpose of your operation. We don't share it with 3rd parties, other than Google / Microsoft for the speech-to-text engine.

Privacy - how are the recordings and results handled?

- transcription service.

Our transcription service is probably the most private and secure transcription service available.

• HIPAA compliant.
• No human in the loop. No passing your recording between PCs, emails, employees, etc.
• Secure encrypted communications (https) with and between our servers.
• Recordings are automatically deleted from our servers as soon as the transcription is done.
• Our contract with Google / Microsoft (our speech engines providers) prohibits them from keeping any audio or results.
• Transcription results are securely kept on our secure database. Only you have access to them - only if you sign in (or provide your secret credentials through the API)
• You may choose to delete the transcription results - once you do - no copy remains on our servers.

- Dictation notepad & extension

For dictation, the recording & recognition - is delegated to and done by the browser (Chrome / Edge) or operating system (Android). So, we never even have access to the recorded audio, and Edge's / Chrome's / Android's (depending the one you use) privacy policy apply here.

The results of the dictation are saved locally on your machine - via the browser's / app's local storage. It never gets to our servers. So, as long as your device is private - your notes are private.

Payments method privacy

The whole payments process is delegated to PayPal / Stripe / Google Pay / Play Store / App Store and secured by these providers. We never receive any of your credit card information.

More generic notes regarding our site, cookies, analytics, ads, etc.

• We may use Google Analytics on our site - which is a generic tool to track usage statistics.
• We use cookies - which means we save data on your browser to send to our servers when needed. This is used for instance to sign you in, and then keep you signed in.
• For the dictation tool - we use your browser's local storage to store your notes, so you can access them later.
• Non premium dictation tool serves ads by Google. Users may opt out of personalized advertising by visiting Ads Settings . Alternatively, users can opt out of a third-party vendor's use of cookies for personalized advertising by visiting https://youradchoices.com/
• In case you would like to upload files to Google Drive directly from Speechnotes - we'll ask for your permission to do so. We will use that permission for that purpose only - syncing your speech-notes to your Google Drive, per your request.

We explain and teach technology, solve tech problems and help you make gadget buying decisions.

5 Ways to Fix Voice-to-Text Not Working on Android

Parth Shah is an evergreen freelance writer covering how-tos, app guides, comparisons, listicles, and troubleshooting guides on Android, iOS, Windows, Mac, and smart TV platforms. He has over five years of experience and has covered 1,500+ articles on GuidingTech and Android Police. In his free time, you will see him binge-watching Netflix shows, reading books, and traveling.

Paras is a staff writer and a published Author, creating engaging SEO-friendly articles that resonate well with readers. Additionally, he diligently updates existing explainers to keep you in the loop with the latest tips and features, helping you make the most of your devices. Over the past 3 years, Paras has been featured on GadgetsToUse, before finally helping the readers at Guiding Tech Media.

• Clean your device’s microphone for better voice input.
• Verify microphone access for Gboard under App info > Permissions > Microphone .
• Enable voice typing under Gboard settings > Voice typing .

Basic Fixes

• Enable system-wide microphone access – Go to Settings > Security and Privacy > enable Microphone access .
• Force stop Gboard and reopen – Open Android settings, locate Gboard among installed apps, open the app, and tap Force Stop .
• Update Gboard – Head to Google Play Store, search Gboard , and tap Update .

Fix 1: Verify Microphone Permissions for the App

If you have denied microphone permissions for a specific app, it can’t detect your voice and convert it into text. Enabling the required permission should fix it.

Step 1: Long-press the app icon to open App info, select Permissions , and tap Microphone .

Step 2: Set Microphone access to Allow only while using the app .

Fix 2: Enable Voice Typing

If voice typing is disabled in your device’s keyboard such as Gboard, you won’t be able to use its Voice-to-text feature. Head to Gboard settings to enable it.

Step 1: Open Gboard, tap its menu in the top left, and press the Settings gear icon.

Step 2: Select Voice typing and enable its toggle. Additionally, turn on the Add punctuation feature if available.

Fix 3: Temporarily Disable Multilingual Typing in Gboard

Turning off the multilingual typing in Gboard temporarily is another effective technique to fix Voice-to-text related issues on Android.

Step 1: Open Gboard settings, tap Languages , and select your preferred language.

Step 2: Toggle off Multilingual typing .

Step 3: Alternatively, tap the Pencil icon in the top right, and select the secondary language to delete it. This will disable the Multilingual typing support.

Fix 4: Download Languages for Better Voice Typing

You should download speech recognition languages when using Gboard to improve voice typing and fix voice-to-text issues in low network strength areas.

Step 1: Open Android settings, tap System , and press Languages .

Step 2: Tap the Settings gear beside Voice input and press Add a language . Download relevant voice packs for a better voice typing experience.

Fix 5: Turn On Personalization for Voice Typing

Enabling personalization for voice typing fine-tunes it and improves voice detection by analyzing your Gboard usage pattern and corrections, fixing existing voice-to-text issues.

Note: This feature may be unavailable on certain Android smartphones. Update Gboard from the Google Play Store.

Step 1: Open Gboard settings, select Voice typing , and tap Manage personalization .

Step 2: Enable the Personalize for you toggle.

Was this helpful?

Last updated on 17 April, 2024

The above article may contain affiliate links which help support Guiding Tech. However, it does not affect our editorial integrity. The content remains unbiased and authentic.

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

The article above may contain affiliate links which help support Guiding Tech. The content remains unbiased and authentic and will never affect our editorial integrity.

DID YOU KNOW

More in Android

Top 11 Ways to Fix Google Chrome Not Working on Mobile Data

7 Best Fixes for Google Slides App Not Working on iPhone and Android

Join the newsletter.

Get Guiding Tech articles delivered to your inbox.

• Read&Write for Windows
• Read&Write for Mac
• Read&Write for Google Chrome
• Read&Write for Windows (Arabic)
• Read&Write for Microsoft Edge
• Read&Write for iOS
• Read&Write for Android
• Remote Learning
• Fluency Tutor for Google
• Articulos De Soporte En Castellano
• OrbitNote for LMS
• Brightspace D2L
• InfiniteCampus
• Using Read&Write with Test Browsers
• Buy Products Online
• Video Tours
• Deployment and Installation guides
• Data & Privacy Policies
• Technical Support Policy
• Terms of Use - UK
• Terms of Use - US
• Terms of Use - Canada
• Terms of Use - Australia & New Zealand
• Terms of Use - Rest of the World
• Open Source Software Providers
• ClaroRead For Windows
• ClaroRead For Mac
• ClaroRead For Chromebook and Chrome Browser
• Claro For iPhone And Apple iPad
• Claro For Android
• Claro Cloud Apps
• Network Installation And Site Licence Support
• Writing Helper
• Global Tasks
• AutoCorrect
• Adding OrbitNote For Canvas Through The Canvas App Center
• Using OrbitNote in D2L Brightspace
• Using OrbitNote With Schoology
• Admin Installation Of Equatio For Canvas
• Teacher/Course Designer Installation Of Equatio For Canvas
• Adding Equatio For Canvas Through The Canvas App Center
• Using Equatio In Canvas With The Rich Text Editor
• Admin Installation Of Equatio LTI For Brightspace D2L
• Enabling The LTI Content Item Messaging Variable In Brightspace
• Admin Installation Of Equatio For Schoology
• What Equatio Features Are Available In Schoology?
• Adding the Equatio LTI to your InfiniteCampus tenant
• Versions Eligible For Technical Support And Out-of-Date Versions No Longer Supported
• Latest Product Versions
• DSA / Access to Work
• Claro Contact Us
• Licence Lookup
• Claro Cloud Service Status
• System Requirements For ClaroRead
• Release Notes
• System Requirements And Download
• Roadmap And Release Notes
• Writing Helper FAQ
• Writing Helper User Guide
• Writing Helper Mac Release Notes
• Writing Helper For Mac - Uninstaller
• Writing Helper Example Projects
• "A client-side exception has occurred"
• Writing Helper Network Installation
• Unable to move, re-order or add new sections/sub-sections in Structure tab
• How to use Global Tasks
• Global Tasks Frequently Asked Questions
• Global Tasks Troubleshooting
• Importing a Mind Map
• Clearing your Site data and Cache
• Mobile notifications not working
• uPAR now has it's own website.
• How do I add educators on the uPAR Website?
• How do I make an Educator an Admin on the uPAR website?
• How do I give a student access to uPAR using a username and password?
• How do I invite students into uPAR by email?
• How do students accept a uPAR invite?
• How do my students take the uPAR protocol?
• How do I add students by Google OU (Organizational Unit) to uPAR?
• How do I remove my Google OU from uPAR?
• Steps to Create a CSV file of students using Google Sheets.
• How do I add students to uPAR using a CSV file?
• How do I import students from Google Classroom into uPAR?
• How do I remove a Google Classroom from uPAR?
• How do I remove a student from uPAR?
• uPAR 1: Overview
• uPAR 2: Giving the Protocol
• uPAR 3: Viewing and Interpreting Data
• uPAR 4: Student Data Talk
• How do I view student data in uPAR?
• How do I change the from and to dates for the uPAR Summary?
• Under All Students what are my options when I go to Print, Download or create a CSV of the Summary Report?
• Under Google Classroom classes what are my options when I go to Print, Download or create a CSV of the Summary Report?
• Under Google OU what are my options to export the Summary Report?
• New uPAR Data sorting options.
• List of uPAR Whitelist items
• I added a student by email but they do not have an info button to the left of their name.
• Steps to select the Legacy Version of uPAR.
• How do I create a Legacy protocol on the uPAR website if I am manually entering in the students.
• Steps to create a Protocol CSV file for Legacy uPAR.(Using Google Sheets)
• How do I edit a Legacy protocol I have created on the uPAR website.
• How do I readminister a Legacy protocol from the uPAR website?
• How can I create a Legacy uPAR protocol using the Add CSV option?
• How do my students login to a Legacy uPAR protocol.
• How can I see the Protocol website/URL and password I created so I can share it with my students.
• How do I view a Legacy uPAR Protocols summary on the uPAR website?
• How do I view the Legacy Cumulative Summary on the uPAR website?
• How can the summary data in Legacy uPAR help me support my students?
• How do I search for a specific students in Legacy uPAR protocol results ?
• How do I archive, unarchive or delete a Protocol on the Legacy uPAR website?
• Locating a students Legacy protocol results if they logged in by email not by protocol name.
• I can't set up a Legacy uPAR assessment for students that have a higher independent reading level then their grade level.
• Is there a way for me to see which passages a student read when they took the Legacy uPAR reading assessment?
• Can more then one Educator be associated with a Legacy uPAR assessment?
• Legacy uPar assessment questions are not matching the audio.
• Does Legacy uPAR have Lexile levels for reading passages and what method does uPAR use to get them to grade level?
• Administrators can now remove stories if they feel the content is against policy.
• Administrators can now view the word count and the text from the stories/passages.
• New Change My Grade option.
• Deprecation of Equatio Server Email
• Simplify AI Disclaimer
• AutoCorrect Licensing and Activation Help
• System Requirements and Downloads
• How To Use AutoCorrect
• AutoCorrect Frequently Asked Questions
• Disable Local Auto Correct
• AutoCorrect Mac Command Line Installation
• AutoCorrect Command Line Install Switches
• Backing Up and Restoring AutoCorrect List and Settings
• Adding Corrections and App Exclusions to AutoCorrect
• Unable to update

How to Fix Read&Write for Windows not Reading in Microsoft Word

If you try to read text in Microsoft Word and it won't work, or if any of Read&Write's tools don't function in Word when they work fine in other apps, this can happen after updates to Office are installed. In this case files used by Read&Write and Microsoft Word can become damaged and need to be repaired.  

To do this, please follow the steps below:

1) Open your Control Panel;  the quickest method to do this is to search for 'control panel' in your Windows Search Bar.

2) In Control Panel, click on 'Programs > Programs and Features'.

3) In the list of programs that appears, find your installation of Microsoft Office and click it to select it.

4) Choose the 'Change' option from the menu at the top of the list.

5) The Office Repair window should now open. Select 'Quick Repair' from the options and then follow the on-screen steps to complete the repair.

6) If this does not resolve the issue, repeat Step 5 but this time perform an ‘Online Repair’.

Once the repair(s) is complete, try again to use the features of Read&Write you were having trouble with in Microsoft Word.

If you still have issues after trying both a Quick and Online Repair, then the next best step is to reinstall Microsoft Office. If you are a technical person (or have access to one), there are also some other fixes you can try:

Delete the normal.dot or normal.dotm document template.

Close Word.

Open Windows Explorer.

Turn on “Show hidden files, folder and drives”.

 “C:\Users\[your username]\AppData\Roaming\Microsoft\Templates”

Delete normal.dotm

Restart Word.

Word may need to re-register with Windows. To do this you will need to open the run command. To do this: Click Start and type ‘run’ in the search box, this should bring up the Run shortcut, click on it.

Depending on your version of office type the following into the run command. If you are using a 64-bit version of Windows, replace “\Program Files\” with “\Program Files (x86)\” "C:\Program Files\Microsoft Office\root\Office16\WINWORD.EXE" /r  

Close Read&Write. Click on Start > All Apps/Programs > TextHelp, and right-click on the Read&Write shortcut. Select “Run as Administrator”. If this does not work, try doing the same for Microsoft Word, and finally for both. Read&Write does not need to be run as an administrator but permissions can be confused on a machine so one or both of Read&Write and Word need to be run elevated. If you find this is the case, try the following steps:

Navigate to Microsoft Word, usually in Program Files > Microsoft Office > Officexx, and called winword.exe. Right-click on winword.exe and select Properties. Open the Security tab.

Navigate to Read&Write, usually in Program Files > TextHelp > Read And Write 12 > Read&Write, and called ReadAndWrite.exe. Right-click on ReadAndWrite.exe and select Properties. Open the Security tab.

Check that the “Group or User names” list is the same for each program. For each entry in the list check that the Permissions list has the same settings for Allow and Deny.

Change the “Read” and “Read & Execute” entries for Read&Write (or Word) so they are the same.

Check your network administrator, if you have one: has anything in Microsoft Word been disabled, for example in the Group Policy Editor or Group Policy Management Console?

The Windows Registry contains keys that must all correctly reference your installation of Microsoft Word. Repairing Microsoft Word may fix your installation of Word but may leave intact erroneous keys – for example, keys from a different version of Word that are not referenced by your current version. One key we have discovered is HKEY_CLASSES_ROOT > TypeLib > {00020905-0000-0000-C000-000000000046}. Child keys of this key reference a special file in Word, the MSWORD.OLB file. If you have any child keys in here that reference a wrong or missing MSWORD.OLB file, then Read&Write will not speak correctly in Word. You can do two things to fix these MSWORD.OLB references:

Search through the child keys of {00020905-0000-0000-C000-000000000046} for any references to an MSWORD.OLB file that is not part of your current installation, or is not actually where it says it is on the computer. For example, if you find a child key that references C:\Program Files (x86)\Microsoft Office\Root\Office16\MSWORD.OLB (the Word 2016 file) but you have Word 2013 installed, then you should delete the child key.

You may also need to add back in the correct reference to the current MSWORD.OLB. You are probably best to use the Office repair operations detailed above. Alternatively, visit the Microsoft Support   website to get additional information/support how to fix this issue.

Run Word in Safe Mode so Add-ins and Extensions are disabled. If it then works this may be indicative of a conflict with an Add-in. Disable Add-ins in Word.

Use Task Manager to check for instances of winword.exe – when Read&Write is running there should be two. Close Word and Read&Write and then check in Task Manager and terminate any other instances. Restart Read&Write and Word and see if the issue is solved.

Try deleting Word add-ins (WLLs), Startup templates and COM Add-ins. Reset user options in Word .

Related Articles

• Installing Microsoft .Net Framework 4.8
• Read&Write Compatibility with Microsoft Word 2021
• Read&Write for Google Chrome Prediction Options
• Deploying Read&Write For Microsoft Edge Using Group Policy
• How to authenticate Read&Write for Google Chrome with an Office 365 account

Contribute to the Microsoft 365 and Office forum! Click  here  to learn more  💡

April 9, 2024

Contribute to the Microsoft 365 and Office forum!

Click  here  to learn more  💡

Word Top Forum Contributors: Stefan Blom  -  Charles Kenyon  -  Suzanne S. Barnhill  -  Doug Robbins - MVP Office Apps & Services (Word)  -  Jim_ Gordon   👍✅

April 17, 2024

Word Top Forum Contributors:

Stefan Blom  -  Charles Kenyon  -  Suzanne S. Barnhill  -  Doug Robbins - MVP Office Apps & Services (Word)  -  Jim_ Gordon   👍✅

• Search the community and support articles
• Microsoft 365 and Office
• Search Community member

Ask a new question

Natural-Sounding Voice in Word Doesn't Work

I'm enjoying the natural-sounding voice feature in Word. See this article:

https://insider.office.com/es-es/blog/new-natural-sounding-voices-come-to-read-aloud

I recently purchased a new PC, and installed 365. (I'm using the same Office 365 account, which has multiple licenses.)

On my new PC, I'm not getting the natural-sounding voice.

What am I doing wrong?

Thank you in advance!

Report abuse

Replies (4) .

• Microsoft Agent |

Hi LA User,

Thanks for posting in the community.

Recently, We received several reports about this issue, and all the customers mentioned these changes and we also can reproduce the same behavior. Based on our test, please kindly see the summary below:

Issue symptom:

Read Aloud voice is changed. Previously it has male or female options.

Now it only has Susan, Hazel, or George's voices sound very robotic for English language.

Environment:

Office for Windows

Cause of the problem:

Refer to Learning Tools in Word ,

Refer to How to download Text-to-Speech languages for Windows 10 , for English different language packages, they have different voices. David, Mark, and Zira are the voices for English (United States).

Workaround:

Use other language packages to see whether there’s any nice voice they like.

Use Word for the Web Immersive Reader instead.

Further Suggestions :

Please send your feedback to Microsoft to improve the language voice.

Feedback in Word: File > Feedback

Feedback in Windows: Use Feedback Hub app.

For more information, we have set a primary thread to continuously update the latest information and solutions. Please click this hyperlink How to restore Word Read Aloud original voice - Microsoft Community for more information. We highly appreciate your understanding and cooperation.

Was this reply helpful? Yes No

Sorry this didn't help.

Great! Thanks for your feedback.

How satisfied are you with this reply?

Thanks for your feedback, it helps us improve the site.

Thanks for your feedback.

Thanks for your response.

Unfortunately, none of this worked. :(

Installing the Australian voice packet didn't do anything.

Also, I don't see a feedback option in the file menu.

If you don't see a feedback option you there are other ways to submit a feedback

Feedback Hub

The “Answers” forum / community is “staffed” mostly by users like yourself, unpaid volunteers. This forum is intended more for “How to do something in application A” type questions. We sometimes can provide work-arounds for problems, but we can’t do anything about underlying problem. If the forum can’t solve your question you need to talk directly to MS, you need to submit a “Feedback” to MS. .

For a couple of years now MS has had a "feedback" process that allows users submit bug reports / problems, suggestions and new feature requests directly to MS developers and managers . You can submit feedbacks from within many MS apps, or you can use the Win10 “Feedback Hub” app. This article describes the different ways you can submit Feedback to MS: https://answers.microsoft.com/en-us/msoffice/forum/all/microsoft-feedback-process/6f850368-8c86-481d-a207-d95b265c4be4 .

There is also a new web based “ Feedback Portal ” in development. It is still labelled “Preview”. It provides functionality similar to the feedback hub. Access to feedbacks submitted in 34 categories. https://feedbackportal.microsoft.com/feedback/

As users we are unpaid quality control testers, we identify bugs, they pick which feedbacks they will work on, and when they will release them, for testing by Insiders and to general public. .

But don't "expect" anything like feedback about your feedback back from MS. .

MS will make fixes and changes on their own schedule. .

They will not provide much, if any, advance notice of what they are working on and/or when it will be released. .

Hint : get a “share” link to your feedback (may take a couple of minutes to process the initial submission of the feedback) and post the link back to this question. That way if someone finds this question they will be able to link directly to your feedback so they can vote for it and add their comments. The more feedback on a subject, the more likely it is to be acted on.

"Feedback" suggesting that they create a "Support / Support" category in the feedback app to collect complaints about poor service in MS Support: https://aka.ms/AAdyeq0 .

PS: also be publicly vocal about your complaints.  Post on social media https://cloudblogs.microsoft.com/industry-blog/microsoft-in-business/media-comm/2017/04/18/microsoft-uses-social-media-create-customer-experience/ .

https://www.facebook.com/pg/Microsoft365/posts/

1 person found this reply helpful

Has this issue been fixed? I just recently started using the text to speech feature and the first time I used it, I had the male and female versions which read the text very nicely. Now it's these super robotic, awful sounding voices.

Question Info

• For education
• Norsk Bokmål
• Ελληνικά
• Русский
• עברית
• العربية
• ไทย
• 한국어
• 中文(简体)
• 中文(繁體)
• 日本語

• Shop all deals
• Free phones
• Smartphones
• Fios Home Internet
• Bring your own device
• Accessories
• Refer a Friend
• Verizon Visa® Card
• Certified pre-owned phones
• Apple iPhone 15 Pro
• Apple iPhone 15
• Samsung Galaxy S24 Ultra
• Google Pixel 8 Pro
• Other phones
• Trade in your device
• Tablets & laptops
• Certified pre-owned watches
• Jetpacks & hotspots
• Shop all accessories
• Phone cases
• Screen protectors
• Tablet accessories
• Chargers & cables
• Phone attachments
• MagSafe compatible
• Verizon accessories
• Shop all watch accessories
• Smart watches
• Shop all plans
• International services
• Connected devices
• Discounts overview
• Mobile + Home
• First responders
• Verizon Forward
• Connected car plans
• Shop all home solutions
• 5G Home Internet
• LTE Home Internet
• Accessories overview
• Cables & connectors
• Networking & Wi-Fi
• TV accessories
• Phone equipment
• 5G Home accessories
• Prepaid overview
• Phone plans
• International plans
• Basic phones
• Mobile hotspots & routers
• Affordable Connectivity Program
• Why Verizon Prepaid
• Disney+, Hulu, ESPN+
• Apple Arcade
• Google Play Pass
• Apple Music
• Xbox All Access
• Services & perks overview
• Entertainment
• Protection & security
• Digital family
• Financial services
• Back to Menu
• Sign in to My Account
• Prepaid instant pay
• Business Log in

Choose your cart

• Mobile solutions
• Home solutions
• My Verizon Member
• Community Member only
• Discussions
• Other Phones
• Speech to text not working on TCL Flip Pro since a...
• Mark as New
• Subscribe to RSS Feed
• Report Inappropriate Content

Basic Phones

• All Community topics

• Getting started / Community FAQs
• Announcements
• How-to videos
• Knowledge Hub
• Introduce Yourself
• Terms of Service and User Guidelines
• Privacy Policy
• Device troubleshooting
• Check network status
• Fios outage page
• Verizon contact numbers

Add thoughts to the community

Start a new topic or ask questions in the open forum.

• New conversation
• Make a suggestion
• Just for fun

Browse discussions within categories

• Accounts & Billing
• Services & perks
• Home Internet & TV
• Support overview
• Return policy
• Community Forums
• Business support
• Download My Verizon App
• Accessibility
• Responsibility
• Verizon Innovative Learning
• Consumer info
• 5G overview
• Innovation Labs
• Apple iPhone 15 Pro Max
• Apple iPhone 15 Plus
• Apple AirPods Max
• Apple Watch Series 9
• Elizabeth James
• Terms & Conditions
• Device Payment Terms & Conditions
• Report a security vulnerability
• Mobile customer agreement
• Radio frequency emissions
• Taxes & surcharges
• Legal notices
• facebook-official
• California Privacy Notice
• Health Privacy Notice
• Open Internet
• Terms & Conditions
• About Our Ads

• International edition
• Australia edition
• Europe edition

Sunak to cite Britain’s ‘sicknote culture’ in bid to overhaul fit note system

Prime minister to express concern about ‘over-medicalising’ normal worries with diagnoses as mental health conditions

Rishi Sunak will today claim Britain is suffering from a “sicknote culture”, as he warns there is a risk of “over-medicalising” normal worries by diagnosing them as mental health conditions.

In a speech on how to reduce people being signed off sick from work, the prime minister will say the government is planning to trial getting “work and health professionals” to issue fit notes, shifting away from GPs carrying out this role.

Sunak will say he is concerned about the increase in long-term sickness since the pandemic, largely driven by mental health conditions with 2.8 million people now “economically inactive”.

In relation to mental health, he will say he would “never dismiss or downplay the illnesses people have”, but also argue that there is a need to be “more honest about the risk of over-medicalising the everyday challenges and worries of life”.

His language echoes that of Mel Stride, the work and pensions secretary, who has claimed that too often doctors “label or medicalise” conditions which in the past were seen as “the ups and downs of life”.

As part of a new drive to overhaul the system, Sunak will say one plank of the reforms will be testing whether responsibility for issuing sicknotes should be shifted from “overstretched” GPs to “specialist work and health professionals who have the dedicated time to provide an objective assessment of someone’s ability to work and the tailored support they need to do so”.

The law was changed last year to allow fit notes to be issued by any doctor, nurse, pharmacist, physiotherapist or occupational therapist in addition to GPs, who have traditionally overseen the system. No 10 last night would not clarify whether the government was planning to widen the criteria further to allow non-medical professionals to issue or change fit notes as well.

Sunak will suggest in his speech that GPs are signing people off sick for work “by default”, with 11m fit notes issued last year, of which 94% assessed people as “not fit for work”. No 10 claimed that the fit note system has “opened the floodgates for millions of people to be written off work and into welfare without getting the right support and treatment they might need to help them stay in work”.

However, extracts of the speech released to the media on Thursday night did not address the issue of NHS mental health capacity, with NHS bosses warning “overwhelmed” services have been unable to cope with a big post-Covid increase in people needing help.

Dr Sarah Hughes, chief executive of the mental health charity Mind, said: “We are deeply disappointed that the prime minister’s speech continues a trend in recent rhetoric which conjures up the image of a ‘mental health culture’ that has ‘gone too far’.

“This is harmful, inaccurate and contrary to the reality for people up and down the country. The truth is that mental health services are at breaking point following years of underinvestment, with many people getting increasingly unwell while they wait to receive support.”

She added: “To imply that it is easy both to be signed off work and then to access benefits is deeply damaging. It is insulting to the 1.9 million people on a waiting list to get mental health support, and to the GPs whose expert judgment is being called into question.”

James Taylor, the director of strategy at the disability charity Scope, said: “Much of the current record levels of inactivity are because our public services are crumbling, the quality of jobs is poor and the rate of poverty among disabled households is growing.”

Employment experts said the number of fit notes issued – 11m last year – has not risen since before the pandemic.

after newsletter promotion

“Fit notes are not driving the increase in economic inactivity. They are not responsible for high levels of worklessness from ill health. Economic inactivity has been driven by people who are already out of work and have been for long periods,” said Tony Wilson, the director of the Institute for Employment Studies.

Plans to reform the fit note system were first flagged by Jeremy Hunt in the autumn statement. A new service called WorkWell is due to be launched this year in 15 areas , which is described as “early-intervention work and health support and assessment”. This will be targeted largely at people with mental health and musculoskeletal conditions, with people who have fit notes attending sessions with “work and health coaches” in an effort to get them to return to employment.

The last Labour government introduced “fit notes” rather than “sicknotes” in 2010 to put a new emphasis on doctors certifying what patients are able to do, rather than what they cannot.

Alison McGovern, Labour’s acting shadow work and pensions secretary, said: “We’ve had 14 Tory years, five Tory prime ministers, seven Tory chancellors, and the result is a record number of people locked out of work because they are sick – at terrible cost to them, to business and to the taxpayer paying billions more in spiralling benefits bills.”

Ruth Rankine, the director of the NHS Confederation’s primary care network, said its members had “long advocated for a review of the fit note process which could be more effectively managed through trained professionals that support people back into work”.

But she added: “The deeper problem isn’t the system – it’s that people are sicker than they were and they have more complex healthcare needs. This is why it is vital the government starts treating investment in the NHS as an explicit tool of economic development and also, that the prime minister should lead a national mission for health improvement to shift the focus from simply treating illness to promoting health and wellbeing.”

• Health policy
• Mental health
• Rishi Sunak

More on this story

Those signed off feeling depressed classed as incapable of work, says Stride

Sunak’s disability benefit plans are familiar culture war fodder

Mental illness costs England £300bn a year, study shows

Disabled people must work from home to do ‘their duty’, says UK minister

Police in England must keep answering mental health calls, charity urges

Ministers’ failures mean disabled people in UK face growing poverty risk – report

20,000 people off work in the UK every month for mental ill health

DWP policy of cold-calling disabled people over benefit claims to end

Record numbers of UK families had benefits capped during lockdown this year

People in 20s more likely to be out of work because of poor mental health than those in early 40s

Most viewed.

Text-to-Speech

Text-to-Speech software reads on-screen text aloud. It helps to reinforce understanding, increase focus and improve retention.

Available with

Text-to-Speech software helps make digital text easier to understand

Text-to-Speech reads on-screen text aloud. Hearing text read aloud reinforces understanding, increases focus and improves retention. This is particularly beneficial for individuals with Dyslexia or ADHD who may find it easier, or more engaging, to process information by listening.

Text-to-Speech is a key feature of Read&Write

The Text-to-Speech functionality within Read&Write reads on-screen text aloud using a naturally sounding voice. Users can even personalise the voice they hear by choosing their favourite voice from a list of options. As the words are being spoken they’re also highlighted. This further supports information processing.

Here’s what our product users say

Being able to have text read aloud is just something I would be lost without. Just to be able to listen to lots of paperwork is great because for me to read it would be very, very difficult.

With Text-to-Speech, employees can work with more confidence

Text-to-Speech offers employees a choice in how they engage with written information.

It helps employees with Dyslexia to process content much quicker. It helps those with ADHD to keep focused. It allows auditory processors to understand in the way that suits them. When employees can understand in their own way, they’re more likely to feel confident. They’re more likely to thrive.

In addition, Text-to-Speech can also give employees a boost of confidence in their written communication. That’s because it’s also a very powerful proofreading tool that can amplify literacy errors. 

Read&Write [Texthelp’s Text-to-Speech software] allows me to listen to emails and documents I’ve written. I can hear errors in my text that I wouldn’t be able to identify just by reading them. This helps me draft high-quality responses and solutions to clients.

Discover our Text-to-Speech software

Supports people who think, learn and work differently. Helps neurodiverse workforces to thrive.

• Universal Design for Learning
• Math and STEM
• Inclusive Education
• Special Education
• Exams and Assessments
• Disabled Students Allowance Hub
• Read&Write
• Inclusive Workplaces
• Neurodiverse Employees
• Creating Accessible Content

Resource centre

• Impact Studies
• Customer Stories

Learn & Support

• Tech Support
• Certification Program
• Toolmatcher
• Accessibility Roadmap

Our approach

• Responsibility
• Infrastructure
• Try Meta AI

RECOMMENDED READS

• 5 Steps to Getting Started with Llama 2
• The Llama Ecosystem: Past, Present, and Future
• Introducing Code Llama, a state-of-the-art large language model for coding
• Meta and Microsoft Introduce the Next Generation of Llama
• Today, we’re introducing Meta Llama 3, the next generation of our state-of-the-art open source large language model.
• Llama 3 models will soon be available on AWS, Databricks, Google Cloud, Hugging Face, Kaggle, IBM WatsonX, Microsoft Azure, NVIDIA NIM, and Snowflake, and with support from hardware platforms offered by AMD, AWS, Dell, Intel, NVIDIA, and Qualcomm.
• We’re dedicated to developing Llama 3 in a responsible way, and we’re offering various resources to help others use it responsibly as well. This includes introducing new trust and safety tools with Llama Guard 2, Code Shield, and CyberSec Eval 2.
• In the coming months, we expect to introduce new capabilities, longer context windows, additional model sizes, and enhanced performance, and we’ll share the Llama 3 research paper.
• Meta AI, built with Llama 3 technology, is now one of the world’s leading AI assistants that can boost your intelligence and lighten your load—helping you learn, get things done, create content, and connect to make the most out of every moment. You can try Meta AI here .

Today, we’re excited to share the first two models of the next generation of Llama, Meta Llama 3, available for broad use. This release features pretrained and instruction-fine-tuned language models with 8B and 70B parameters that can support a broad range of use cases. This next generation of Llama demonstrates state-of-the-art performance on a wide range of industry benchmarks and offers new capabilities, including improved reasoning. We believe these are the best open source models of their class, period. In support of our longstanding open approach, we’re putting Llama 3 in the hands of the community. We want to kickstart the next wave of innovation in AI across the stack—from applications to developer tools to evals to inference optimizations and more. We can’t wait to see what you build and look forward to your feedback.

Our goals for Llama 3

With Llama 3, we set out to build the best open models that are on par with the best proprietary models available today. We wanted to address developer feedback to increase the overall helpfulness of Llama 3 and are doing so while continuing to play a leading role on responsible use and deployment of LLMs. We are embracing the open source ethos of releasing early and often to enable the community to get access to these models while they are still in development. The text-based models we are releasing today are the first in the Llama 3 collection of models. Our goal in the near future is to make Llama 3 multilingual and multimodal, have longer context, and continue to improve overall performance across core LLM capabilities such as reasoning and coding.

State-of-the-art performance

Our new 8B and 70B parameter Llama 3 models are a major leap over Llama 2 and establish a new state-of-the-art for LLM models at those scales. Thanks to improvements in pretraining and post-training, our pretrained and instruction-fine-tuned models are the best models existing today at the 8B and 70B parameter scale. Improvements in our post-training procedures substantially reduced false refusal rates, improved alignment, and increased diversity in model responses. We also saw greatly improved capabilities like reasoning, code generation, and instruction following making Llama 3 more steerable.

*Please see evaluation details for setting and parameters with which these evaluations are calculated.

In the development of Llama 3, we looked at model performance on standard benchmarks and also sought to optimize for performance for real-world scenarios. To this end, we developed a new high-quality human evaluation set. This evaluation set contains 1,800 prompts that cover 12 key use cases: asking for advice, brainstorming, classification, closed question answering, coding, creative writing, extraction, inhabiting a character/persona, open question answering, reasoning, rewriting, and summarization. To prevent accidental overfitting of our models on this evaluation set, even our own modeling teams do not have access to it. The chart below shows aggregated results of our human evaluations across of these categories and prompts against Claude Sonnet, Mistral Medium, and GPT-3.5.

Preference rankings by human annotators based on this evaluation set highlight the strong performance of our 70B instruction-following model compared to competing models of comparable size in real-world scenarios.

Our pretrained model also establishes a new state-of-the-art for LLM models at those scales.

To develop a great language model, we believe it’s important to innovate, scale, and optimize for simplicity. We adopted this design philosophy throughout the Llama 3 project with a focus on four key ingredients: the model architecture, the pretraining data, scaling up pretraining, and instruction fine-tuning.

Model architecture

In line with our design philosophy, we opted for a relatively standard decoder-only transformer architecture in Llama 3. Compared to Llama 2, we made several key improvements. Llama 3 uses a tokenizer with a vocabulary of 128K tokens that encodes language much more efficiently, which leads to substantially improved model performance. To improve the inference efficiency of Llama 3 models, we’ve adopted grouped query attention (GQA) across both the 8B and 70B sizes. We trained the models on sequences of 8,192 tokens, using a mask to ensure self-attention does not cross document boundaries.

Training data

To train the best language model, the curation of a large, high-quality training dataset is paramount. In line with our design principles, we invested heavily in pretraining data. Llama 3 is pretrained on over 15T tokens that were all collected from publicly available sources. Our training dataset is seven times larger than that used for Llama 2, and it includes four times more code. To prepare for upcoming multilingual use cases, over 5% of the Llama 3 pretraining dataset consists of high-quality non-English data that covers over 30 languages. However, we do not expect the same level of performance in these languages as in English.

To ensure Llama 3 is trained on data of the highest quality, we developed a series of data-filtering pipelines. These pipelines include using heuristic filters, NSFW filters, semantic deduplication approaches, and text classifiers to predict data quality. We found that previous generations of Llama are surprisingly good at identifying high-quality data, hence we used Llama 2 to generate the training data for the text-quality classifiers that are powering Llama 3.

We also performed extensive experiments to evaluate the best ways of mixing data from different sources in our final pretraining dataset. These experiments enabled us to select a data mix that ensures that Llama 3 performs well across use cases including trivia questions, STEM, coding, historical knowledge, etc.

Scaling up pretraining

To effectively leverage our pretraining data in Llama 3 models, we put substantial effort into scaling up pretraining. Specifically, we have developed a series of detailed scaling laws for downstream benchmark evaluations. These scaling laws enable us to select an optimal data mix and to make informed decisions on how to best use our training compute. Importantly, scaling laws allow us to predict the performance of our largest models on key tasks (for example, code generation as evaluated on the HumanEval benchmark—see above) before we actually train the models. This helps us ensure strong performance of our final models across a variety of use cases and capabilities.

We made several new observations on scaling behavior during the development of Llama 3. For example, while the Chinchilla-optimal amount of training compute for an 8B parameter model corresponds to ~200B tokens, we found that model performance continues to improve even after the model is trained on two orders of magnitude more data. Both our 8B and 70B parameter models continued to improve log-linearly after we trained them on up to 15T tokens. Larger models can match the performance of these smaller models with less training compute, but smaller models are generally preferred because they are much more efficient during inference.

To train our largest Llama 3 models, we combined three types of parallelization: data parallelization, model parallelization, and pipeline parallelization. Our most efficient implementation achieves a compute utilization of over 400 TFLOPS per GPU when trained on 16K GPUs simultaneously. We performed training runs on two custom-built 24K GPU clusters . To maximize GPU uptime, we developed an advanced new training stack that automates error detection, handling, and maintenance. We also greatly improved our hardware reliability and detection mechanisms for silent data corruption, and we developed new scalable storage systems that reduce overheads of checkpointing and rollback. Those improvements resulted in an overall effective training time of more than 95%. Combined, these improvements increased the efficiency of Llama 3 training by ~three times compared to Llama 2.

Instruction fine-tuning

To fully unlock the potential of our pretrained models in chat use cases, we innovated on our approach to instruction-tuning as well. Our approach to post-training is a combination of supervised fine-tuning (SFT), rejection sampling, proximal policy optimization (PPO), and direct preference optimization (DPO). The quality of the prompts that are used in SFT and the preference rankings that are used in PPO and DPO has an outsized influence on the performance of aligned models. Some of our biggest improvements in model quality came from carefully curating this data and performing multiple rounds of quality assurance on annotations provided by human annotators.

Learning from preference rankings via PPO and DPO also greatly improved the performance of Llama 3 on reasoning and coding tasks. We found that if you ask a model a reasoning question that it struggles to answer, the model will sometimes produce the right reasoning trace: The model knows how to produce the right answer, but it does not know how to select it. Training on preference rankings enables the model to learn how to select it.

Building with Llama 3

Our vision is to enable developers to customize Llama 3 to support relevant use cases and to make it easier to adopt best practices and improve the open ecosystem. With this release, we’re providing new trust and safety tools including updated components with both Llama Guard 2 and Cybersec Eval 2, and the introduction of Code Shield—an inference time guardrail for filtering insecure code produced by LLMs.

We’ve also co-developed Llama 3 with torchtune , the new PyTorch-native library for easily authoring, fine-tuning, and experimenting with LLMs. torchtune provides memory efficient and hackable training recipes written entirely in PyTorch. The library is integrated with popular platforms such as Hugging Face, Weights & Biases, and EleutherAI and even supports Executorch for enabling efficient inference to be run on a wide variety of mobile and edge devices. For everything from prompt engineering to using Llama 3 with LangChain we have a comprehensive getting started guide and takes you from downloading Llama 3 all the way to deployment at scale within your generative AI application.

A system-level approach to responsibility

We have designed Llama 3 models to be maximally helpful while ensuring an industry leading approach to responsibly deploying them. To achieve this, we have adopted a new, system-level approach to the responsible development and deployment of Llama. We envision Llama models as part of a broader system that puts the developer in the driver’s seat. Llama models will serve as a foundational piece of a system that developers design with their unique end goals in mind.

Instruction fine-tuning also plays a major role in ensuring the safety of our models. Our instruction-fine-tuned models have been red-teamed (tested) for safety through internal and external efforts. ​​Our red teaming approach leverages human experts and automation methods to generate adversarial prompts that try to elicit problematic responses. For instance, we apply comprehensive testing to assess risks of misuse related to Chemical, Biological, Cyber Security, and other risk areas. All of these efforts are iterative and used to inform safety fine-tuning of the models being released. You can read more about our efforts in the model card .

Llama Guard models are meant to be a foundation for prompt and response safety and can easily be fine-tuned to create a new taxonomy depending on application needs. As a starting point, the new Llama Guard 2 uses the recently announced MLCommons taxonomy, in an effort to support the emergence of industry standards in this important area. Additionally, CyberSecEval 2 expands on its predecessor by adding measures of an LLM’s propensity to allow for abuse of its code interpreter, offensive cybersecurity capabilities, and susceptibility to prompt injection attacks (learn more in our technical paper ). Finally, we’re introducing Code Shield which adds support for inference-time filtering of insecure code produced by LLMs. This offers mitigation of risks around insecure code suggestions, code interpreter abuse prevention, and secure command execution.

With the speed at which the generative AI space is moving, we believe an open approach is an important way to bring the ecosystem together and mitigate these potential harms. As part of that, we’re updating our Responsible Use Guide (RUG) that provides a comprehensive guide to responsible development with LLMs. As we outlined in the RUG, we recommend that all inputs and outputs be checked and filtered in accordance with content guidelines appropriate to the application. Additionally, many cloud service providers offer content moderation APIs and other tools for responsible deployment, and we encourage developers to also consider using these options.

Deploying Llama 3 at scale

Llama 3 will soon be available on all major platforms including cloud providers, model API providers, and much more. Llama 3 will be everywhere .

Our benchmarks show the tokenizer offers improved token efficiency, yielding up to 15% fewer tokens compared to Llama 2. Also, Group Query Attention (GQA) now has been added to Llama 3 8B as well. As a result, we observed that despite the model having 1B more parameters compared to Llama 2 7B, the improved tokenizer efficiency and GQA contribute to maintaining the inference efficiency on par with Llama 2 7B.

For examples of how to leverage all of these capabilities, check out Llama Recipes which contains all of our open source code that can be leveraged for everything from fine-tuning to deployment to model evaluation.

What’s next for Llama 3?

The Llama 3 8B and 70B models mark the beginning of what we plan to release for Llama 3. And there’s a lot more to come.

Our largest models are over 400B parameters and, while these models are still training, our team is excited about how they’re trending. Over the coming months, we’ll release multiple models with new capabilities including multimodality, the ability to converse in multiple languages, a much longer context window, and stronger overall capabilities. We will also publish a detailed research paper once we are done training Llama 3.

To give you a sneak preview for where these models are today as they continue training, we thought we could share some snapshots of how our largest LLM model is trending. Please note that this data is based on an early checkpoint of Llama 3 that is still training and these capabilities are not supported as part of the models released today.

We’re committed to the continued growth and development of an open AI ecosystem for releasing our models responsibly. We have long believed that openness leads to better, safer products, faster innovation, and a healthier overall market. This is good for Meta, and it is good for society. We’re taking a community-first approach with Llama 3, and starting today, these models are available on the leading cloud, hosting, and hardware platforms with many more to come.

Try Meta Llama 3 today

We’ve integrated our latest models into Meta AI, which we believe is the world’s leading AI assistant. It’s now built with Llama 3 technology and it’s available in more countries across our apps.

You can use Meta AI on Facebook, Instagram, WhatsApp, Messenger, and the web to get things done, learn, create, and connect with the things that matter to you. You can read more about the Meta AI experience here .

Visit the Llama 3 website to download the models and reference the Getting Started Guide for the latest list of all available platforms.

You’ll also soon be able to test multimodal Meta AI on our Ray-Ban Meta smart glasses.

As always, we look forward to seeing all the amazing products and experiences you will build with Meta Llama 3.

Our latest updates delivered to your inbox

Subscribe to our newsletter to keep up with Meta AI news, events, research breakthroughs, and more.

Join us in the pursuit of what’s possible with AI.

Product experiences

Foundational models

Latest news

Meta © 2024