Voice transcription

Standalone transcription converts speech to text outside of chat: live streaming from your microphone, batch processing of audio files, or a short in-browser recording. This page documents the three paths, required authentication, and how to manage language and model settings for production-quality notes and compliance-friendly workflows.

Audience — Users who need accurate transcripts for meetings, interviews, dictation, or accessibility—not only chat-based Q&A.

In this article

Authentication and why a valid access token is required
Comparison of live, upload, and clip recording flows
Configurable parameters: language, model, sample rate
Export, copy, and session hygiene
Privacy, browser permissions, and troubleshooting

Access and prerequisites

Open Transcription from the dashboard. The service uses your access token from a signed-in session; without it, the client cannot open a secure channel to the transcription backend.

Browser permissions

Live and clip flows require microphone access. Users must click Allow when prompted. Corporate browsers with locked-down media permissions may block recording entirely—document an approved browser profile for your org.

The three transcription paths

Method	When to use it	Behavior summary
Live (microphone stream)	Real-time meetings, dictation, long-form capture.	Interim text while speaking; final segments append to the transcript. Stop the session to close the connection.
File upload	Pre-recorded interviews, podcasts, voice memos.	Upload completes; server returns a transcript; text is merged into the editor when done.
Record clip	Quick capture without maintaining a long live session.	Browser records audio, then sends it as a file for the same transcription path as upload.

Transcription page with live, upload, and record options.

Live transcription in detail

Starting live mode opens a streaming connection. Status indicators (connecting, listening, reconnecting, closed) reflect the gateway state. Interim results may appear while you speak; finalized text is appended to the running document. If the connection drops, the UI may attempt reconnection depending on implementation—watch status messages and restart if the session stays closed.

Operational tips

Use a quiet environment or directional mic for best accuracy.
Speak clearly; avoid overlapping speakers if the engine is single-channel.
End the session explicitly when finished to release the microphone.

Upload and recorded clips

For uploads, choose supported audio formats accepted by the file picker. For in-browser recording, the implementation uses the device’s default input device; stop recording before transcribing so the blob is complete. Both paths run through the same server-side transcription with punctuation and smart formatting when enabled.

Language, model, and audio settings

Before starting, set language, model (for example a Nova-class engine where applicable), and sample rate to match your capture hardware when relevant. Mismatched sample rates can reduce quality; consult your audio engineer for broadcast or music-heavy sources.

Consistency

Standardize language codes and model names in your team wiki so support tickets reference the same configuration every time.

Working with the transcript

Use Copy to place the full text on the clipboard for email, docs, or tickets. Use Download to save a timestamped plain-text file for archival. Clear the editor between sessions when you need a clean slate; avoid mixing unrelated recordings in one document for audit clarity.

Privacy and data handling

Treat transcripts like any sensitive business record. Follow your company’s retention policy. Do not stream highly confidential content on untrusted networks; use uploads over secure corporate channels when required by policy.

Troubleshooting

Problem	Remediation
Login required / token error	Sign in again; ensure local storage is not cleared aggressively by policy.
Microphone not available	Check OS privacy settings; try HTTPS; verify no other app holds the device.
Empty or partial transcript	Check language setting; retry with cleaner audio; verify file not corrupted.