Talk to SynapseX with voice
Long prompts are faster to say than to type. This page gets you dictating into the SynapseX Chat composer in whatever language you speak, and then into a hands-free back-and-forth where the assistant reads its answers aloud.
Dictate a message
Click the microphone button in the composer and start talking. The browser records the audio, sends it to SynapseX for transcription, and drops the resulting text into the message box.
Transcription happens server-side with automatic language detection — you do not select a language first, and you can switch languages mid-sentence. A transcription backend is configured in production, so this is the path you normally get.
If server transcription is unavailable, the app falls back to your browser’s built-in Web Speech API. That fallback is single-language and only exists in some browsers, so if you land on it, dictation may be English-only or unavailable.
Try it: dictate a long prompt
- Click the microphone.
- Say the whole prompt out loud, including the constraints — for example: “Review the attached component for accessibility problems. Focus on keyboard navigation and colour contrast. Give me a numbered list, worst first.”
- Click the microphone again to stop.
- Read the transcript in the message box and edit it — fix names, add a code fence, remove filler.
- Press Enter.
Editing before sending is the point. Dictation gets a hundred words into the box in fifteen seconds; you spend the next fifteen making them precise.
Hands-free voice mode
The floating voice button starts a continuous loop: it listens, sends what you said, speaks the reply out loud, and then listens again for your next turn. This runs on your browser’s speech APIs.
Stop it with the same floating button — leaving it running keeps the microphone open.
Requirements and what can go wrong
| Situation | What happens | What to do |
|---|---|---|
| First use | The browser asks for microphone permission | Allow it; a blocked microphone silently records nothing |
| Very long recording | Requests are capped at 25 MB of audio | Record in shorter passes |
| Unsupported browser | Hands-free mode and the browser fallback need browser speech APIs | Use a browser that supports them, or type |
| No sound on playback | The tab may be muted, or autoplay blocked | Unmute the tab and click once in the page |
Voice settings
Settings > Voice has a Language selector for reading and dictation, alongside style and speed controls. If you want the browser fallback and spoken replies to use English, set the language there to English (US). Server-side transcription detects your language automatically regardless of this setting.
Related
- Files and documents — audio files can also be attached to a message.
- Conversations and projects — a dictated chat is a normal chat, so it searches and exports the same way.