Talk to it instead of typing

Voice is two independent halves. Talking to it is speech recognition (our ASRx module), done by your browser or by a transcription service. It talking back is speech synthesis (Speex), done by one of five voice engines. They are configured separately and they fail separately, so one can work while the other does not.

Dictating into the composer

Ctrl + M starts and stops the traditional microphone; the button lights up and the status bar swaps in these while it listens:

Key	While the microphone is live
Enter	Mic · Send
Ctrl + Enter	Mic · Beam
Esc	Mic · Stop

Words appear in the composer as you speak and can be edited before sending, so dictation mixes with typing. A continuation mode re-arms the microphone after each send.

Ramble

Ramble records one take of up to 60 minutes and transcribes the whole file with your own Deepgram or OpenAI key free. The audio stays in this browser; the transcripts sync on Pro pro. Where to start a take, what a reload costs, and what is kept: an hour of speech, transcribed whole.

Holding a spoken conversation

A call is recognition, model generation, and synthesis - three phases in turn. Latency is higher than a realtime voice model; in exchange the call runs on the chat models and the data you already have. Switch to the Calls app, or press the call button in a chat or on an open persona - the call carries that chat's context, or starts with none.

The calls app reports on three things before it dials:

The chat is empty. Calls are effective when the caller has context.
Speech recognition should be good to go.
Voice synthesis should be ready.

all three green, in the footer:   Ready, Set, Call

In the call: Push To Talk while you speak, or leave it listening. Two spoken commands are built in - Stop. makes the assistant hold its answer, Goodbye. ends the call. Ctrl + M still toggles the microphone.

Starting to talk stops the assistant talking. The microphone and the speaker are arbitrated over one audio resource, and the microphone wins. Hanging up cuts playback the same way.

Choosing a voice - synthesis

Five engines synthesize speech, configured in Settings: some run on the LLM keys you already added, some take a key of their own. One engine is active at a time, or Auto picks.

Engine	Configuration	What you can change
Your browser's own voices	Just works	Voice, speed, pitch
ElevenLabs	That vendor's key	Model, voice
Inworld	That vendor's key	Model, voice, speed
OpenAI	Your existing OpenAI key, or a different one	Model, voice, speed - and an instruction, on the Expressive model
A local speech server	Its network URL	Model, backend

Each engine ships with a default voice, and the preview button reads the same line every time - Hello, this is my voice. A voice is stored as an engine plus its settings.

Pin a voice to a persona and every chat with that persona speaks in it, including the line it opens a call with: what a persona is. Outside calls replies are silent unless you ask: auto-speak offers off (the default), the first line, a briefing, or everything. One reply can always be spoken from its message menu's Speak.

All five engines exist in every version. OpenAI voices run on the key your OpenAI service already carries; ElevenLabs and Inworld take their own vendor key (optional services).

Where the audio goes, and what is kept

Feature	Where the audio goes
the microphone button	your browser turns speech into text; Big-AGI receives the text, never audio
Ramble	audio in this browser's storage on this device, then straight to the transcription service you configured
synthesis	the text goes to the voice engine you picked

Whether recognition runs on the device or at the browser vendor's service is the browser's choice, not Big-AGI's. A Ramble entry can be replayed or transcribed again on the device that holds its audio. Deleting the entry deletes the audio with it. The full account of what leaves your machine, and by which route: what leaves your browser.

Browsers and devices

Browser	What happens
Chrome · Edge · Safari	recognition works
Brave	blocked - Speech recognition is not supported in Brave. Please use Chrome, Edge, or Safari.
iOS	a warning rather than a block - Not tested on this browser/device.
no recognition at all	Speech recognition API is not available in this browser.

Brave is blocked deliberately, because it exposes the interface and then returns nothing. Synthesis has no such restriction: any browser can play the audio back.

Troubleshoot - audio errors

What you see	What it means	What to do
Microphone access blocked by the user. Enable it in your browser settings to use speech recognition.	The site permission is denied.	Allow the microphone for this site, then reload.
Speech Recognition permission denied. Check your System Settings.	The operating system is blocking the browser. On macOS Speech Recognition is a permission separate from Microphone, so granting the microphone does not grant it.	System Settings -> Privacy & Security -> Speech Recognition, tick your browser, then quit and reopen it. That pane's path is as documented in September 2024.
Audio capture failed (...). Please try again.	No usable input device.	Check the input device, and that nothing else holds it.
Network communication required to complete the service, but failed.	Browser recognition could not reach its service.	Retry on a stable connection.
Issue starting the speech recognition.	A second start was requested before the first stopped.	Wait a moment and press Ctrl + M again.
Nothing is spoken back	Autoplay is blocked until you interact with the page.	Click once in the page, then retry.

Related

• An hour of speech, transcribed whole
• What a persona is
• What leaves your browser
• Attachments and outputs
• Optional services