Some websites use speech recognition to let you talk instead of type, or to automatically add captions to audio and video that don't have them. In Firefox, speech recognition happens entirely on your device.
Firefox’s built-in speech recognition never sends your audio to Mozilla or to any server. The website only receives the resulting text. Once the required AI model is downloaded, speech recognition can also work without an internet connection.
What is speech recognition in Firefox?
Speech recognition lets a website turn audio into text,whether from your microphone, a video call, a video, or a podcast,using a web standard called the Web Speech API.
Some browsers and services do this by sending your audio to remote servers for processing. Firefox instead downloads an AI model to your device and does all transcription locally.
When a website uses Firefox’s built-in speech recognition, your audio never leaves your device.
Turn speech recognition on or off
You control whether websites are allowed to use speech recognition from Firefox's AI Controls:
- Click the menu button ☰ and select .
- Select in the left panel.
- Under , find .
- Choose an option from the drop-down menu.
- Enabled: Websites can use speech recognition. Firefox asks for your permission before downloading a new model. If the required model is already installed, websites can use it without asking for permission again.
- Blocked: Websites can't use speech recognition. Any speech recognition model already downloaded is removed from your device.
Allow a website to use speech recognition
If a website needs a speech recognition model that isn't already installed, Firefox will ask for your permission before downloading it. You can either continue or decline. If you decline, the website won't be able to use Firefox's built-in speech recognition until you approve the download.
The model Firefox downloads depends on the language the website requests:
- English: Approximately 140 MB
- Other supported languages: A larger multilingual model of approximately 785 MB.
Each model only needs to be downloaded once. After that, it stays on your device and is used by any website that asks for speech recognition.
Supported languages
English uses a dedicated model for English and a multilingual model for other supported languages.
The multilingual model supports around 40 languages, including Spanish, French, German, Italian, Portuguese, Dutch, Turkish, Russian, Arabic, Hindi, Japanese, Korean, Vietnamese, Ukrainian, Polish, Swedish, Czech, and Chinese (Mandarin).
Recognition quality varies by language, and Mozilla is working to improve it over time.
Manage or remove the speech recognition model
Once downloaded, the speech recognition model is listed alongside Firefox's other on-device AI models:
- Click the menu button ☰ and select .
- Select from the menu.
On this page, you can:
- Review models: See which models are downloaded, how large they are, and what they're used for.
- Remove models: Useful if you need to free up space or no longer use speech recognition.
If a website asks to use speech recognition again later, Firefox will prompt you to download the model again.
You can read more in On-device AI models in Firefox.
Troubleshoot speech recognition
If a website's speech recognition prompt never appears, or a site says the feature isn't available:
- Check your AI Controls settings. Go to , select , and make sure is set to .
- Try the feature again. If you declined the model download the first time a site asked, revisit the site and try the feature again; Firefox will prompt you again.
- Check microphone permissions. If the site is transcribing your voice, check that its microphone permission is allowed. This is separate from the speech recognition setting; Firefox needs both.
- Check your device's performance. Speech recognition runs on your device and can use significant resources. Firefox uses your computer’s graphics processor (GPU) when it can and your main processor (CPU) otherwise. On older or less powerful computers, it may run more slowly.
Frequently asked questions
Does Firefox send my audio to a server?
Not when a website uses Firefox’s built-in speech recognition. Firefox processes your audio locally and doesn’t send it to Mozilla or to the website. The website only receives the resulting text. Some websites use their own speech recognition services instead (see below).
Do I need to download anything?
Yes. The first time a website uses speech recognition, Firefox needs to download a model. You'll be asked to approve the download before anything is installed.
Can I use speech recognition without an internet connection?
Yes. Once the model is downloaded, transcription happens on your device and doesn't require an internet connection.
Can I remove the model to save space?
Yes. You can remove it from under .
If a website asks to use speech recognition again afterward, Firefox will prompt you to download it again.
If I block speech recognition, does that stop websites from transcribing my audio?
Not always. This setting controls Firefox's built-in, on-device speech recognition. A website can instead use its own speech recognition service: if you've granted it microphone access, it can record your audio and send it to its own servers for transcription.
In that case, your audio leaves your device and is handled according to that website's privacy policy,not by Firefox. Firefox's on-device speech recognition lets websites offer these features without your audio ever leaving your device.
What if a website asks to use speech recognition and I decline?
The website won't be able to transcribe speech. You can change your mind later, either by accepting the prompt the next time the site asks, or through the Speech recognition setting in AI Controls.
