Live transcript
Microphone starts only when you choose
Your words will appear here
Start a recording, or open the prepared demo to see speaker turns and mixed-language text.
Turn speech into readable text while you are still talking. The browser workbench above can listen through your microphone, display partial and final recognition results, track elapsed time, and let you copy or download the finished text. It is designed for quick notes and draft transcripts without an upload step.
Live transcription is browser-based speech-to-text that converts microphone audio into partial and final text while you speak. Choose the closest recognition language, press Start recording, grant microphone access, and watch the transcript update. Browser support, audio quality, accents, names, and background noise all affect the result, so important transcripts still need human review.
Muse Voice Transcribe is an independent browser tool and is not affiliated with Meta. This page describes the behavior you can test in the workbench above, not unpublished model capabilities.
Last updated:
Select “Open the prepared demo instead” in the workbench to load a fixed four-turn live transcription conversation. The sample is deliberately predictable: it contains two speaker labels, four timestamps, English text, and one mixed Chinese-English turn. You can copy it or download the same content as a TXT file. It is a product sample rather than a customer recording, benchmark, or accuracy claim.
Speaker A [00:04]: Thanks for joining. Let us start with the interview questions. Speaker B [00:09]: 当然，我已经准备好了。The first topic is the launch timeline. Speaker A [00:15]: Great. I will keep the transcript open while we talk. Speaker B [00:21]: Perfect. We can review the final text before sharing it.
The prepared output demonstrates the transcript format. In microphone mode, the browser returns speech-recognition results as they become available and labels the active voice as “You.” It does not automatically assign multiple speakers in this browser-only mode. If the browser does not expose speech recognition, the workbench reports that limitation and leaves the prepared example available.
| Fact | Value |
|---|---|
| Prepared transcript | 4 turns / 21 seconds |
| Recognition languages | 5 choices |
| Live speaker label | You |
| Prepared speaker labels | Speaker A / Speaker B |
| Export | TXT only |
| Browser compatibility | Not published |
| Recognition latency | Not measured |
| Audio recording library | Not provided |
Source: the visible workbench and prepared transcript on this page. Verified September 3, 2026.
The page keeps the capture workflow intentionally small. Each control maps to a concrete step you can verify in the interface rather than a promised feature hidden behind an account.
Recording starts only after you press the button and approve the browser permission. Speech is returned as partial or final text by the recognition engine available in your browser. The page does not silently activate the microphone on load.
Choose English (US), English (UK), Mandarin Chinese, Spanish, or French before recording. This setting tells the browser which language to expect; it is not automatic language detection, translation, or a guarantee that mixed-language speech will be recognized equally well.
Pause a session when the conversation stops, finish when you are ready to review, or begin a new session to clear the current text. The elapsed timer gives each session a simple time reference without pretending to link the transcript to a stored audio recording.
Once text is available, copy it to another document or download a plain-text file. TXT is useful for notes, search, and lightweight editing, but this page does not currently create subtitles, DOCX files, summaries, or a shareable recording.
A short setup check produces better results than starting immediately in a noisy room. Use this sequence for a meeting, interview, lecture, or voice note.
Pick the language that best matches the main speaker. If a session switches languages, finish the current capture, change the setting, and start again. The prepared sample can help you inspect the output layout before granting microphone access.
Press Start recording and respond to the browser permission prompt. Speak close to the microphone at a steady pace. A quiet room and one person speaking at a time usually produce a cleaner draft than distant audio or overlapping conversation.
Partial recognition may change as the engine receives more context. Use the display as a live reference, not as a final record. If a name, number, or key decision looks wrong, make a note to verify it against the source immediately after the session.
Stop the session, read the complete transcript, correct critical details in your destination document, then copy the text or download the TXT file. Keep the original recording separately when the exact wording matters, because this workbench does not retain one for comparison.
Live transcription is most useful when seeing a draft immediately is more valuable than waiting for a polished post-production transcript.
Keep a running text draft beside a short planning call, stand-up, or review. Afterward, extract decisions and action items manually and confirm ownership with the participants. The transcript supports note-taking; it does not replace an agreed meeting record.
Follow answers in text while an interview is happening and mark passages that deserve a follow-up question. Use a separate consented recording when accurate quotations are required, then verify every quote, proper noun, and number before publication.
Capture a lecture outline, a spoken first draft, or a visual text aid for a supported browser. Recognition latency and quality vary, so the page should not be treated as certified captioning, emergency communication, or an accommodation guaranteed to work on every device.
Real-time output is useful only when its boundaries are visible. Check these four points before relying on a transcript.
Noise, distance, echo, fast speech, accents, specialist terms, proper names, numbers, and overlapping voices can all introduce errors. The page does not publish a universal word-error rate because the browser engine and recording conditions are outside a single controlled benchmark.
The microphone workflow depends on browser media permissions and a compatible speech-recognition interface. Some browsers or managed devices may not provide it. When live capture is unavailable, use the prepared transcript to inspect the interface rather than assuming audio is being processed.
This page does not claim that recognition always stays on-device. A browser or its speech service may process audio according to its own implementation. Check browser permissions and the site privacy policy, obtain participant consent, and avoid sensitive material when those terms are unsuitable.
Live browser results are labeled “You.” The two-speaker labels in the prepared example demonstrate a readable transcript format, not automatic identification during your own session. Use the dedicated speaker diarization guide to understand that distinction before planning a multi-person workflow.
The live workflow depends on browser APIs. These public specifications explain the permission and recognition interfaces referenced on this page.
Yes. You can open the prepared transcript or try the microphone controls on this page without an account. Microphone capture still depends on browser support and your permission settings. The page does not promise unlimited hosted processing, storage, or features that are not visible in the workbench.
The visible workbench requests a microphone stream for browser speech recognition and creates transcript text; it does not provide an audio recording download or a recording library. Do not infer that recognition is always on-device. Review your browser behavior and the site privacy policy before handling confidential speech.
Speech engines often return an interim guess and revise it when later words provide more context. That is normal for real-time transcription. Wait for final text when possible and review the complete output after stopping, especially around names, numbers, abbreviations, and punctuation.
Not in the browser microphone mode shown here. It labels the current speech as “You.” The prepared example displays Speaker A and Speaker B only to show the intended output format. Automatic speaker attribution requires a diarization-capable processing workflow and should be evaluated separately.
No. This workbench currently captures a supported browser microphone and exports plain TXT. It does not advertise audio or video upload, SRT or VTT subtitles, translation, automatic summaries, or a public transcription API. Use only the controls that are present on the page.
Read the speaker diarization page when participant labels matter, use the meeting transcription page for a meeting-specific review checklist, or open the Muse Voice Transcribe overview for the broader product context.
Use live transcription by choosing a language, pressing Start recording, and speaking a short test sentence. Review the output before using the workbench for a longer session.
Return to the workbench