Speech Transcription
The Speech Transcription analysis automatically converts a participant's spoken audio into a written transcript. It runs entirely on the research platform's server — no external accounts or internet connection are required.
Overview
After a participant completes a session and their recording is processed, the platform automatically starts transcribing the audio. The result is a time-aligned transcript you can read alongside the recording, export as a plain text file, or download as a timed subtitle file (VTT) for frame-accurate review.
What it produces
- A full written transcript of the participant's speech.
- Timestamped segments: each portion of speech is tagged with its start and end time in the recording, so you can jump to any moment directly in the review interface.
- Optional word-level timing: when enabled, each individual word carries its own start and end time, giving you finer alignment for analysis.
Configuration
When you add Speech Transcription to a study you can choose:
- Language: the primary language spoken by participants (for example, English or French). Selecting the correct language improves accuracy.
Where to find the results
Results appear in the Analysis tab of any participation that has a completed transcription. The transcript is displayed as a scrollable list of timed segments. Clicking a segment jumps the recording playback to that moment.
You can also find the transcription in the Timeline tab as a block track, where each speech segment is drawn as a colored bar aligned to the recording.
Exporting
From the participation download panel you can export:
analyses/speech_transcription.json— the full result file in JSON format, including all segments and their timestamps.transcription/speech_transcription.vtt— a WebVTT subtitle file with one cue per segment, suitable for playback in any video player.transcription/speech_transcription.csv— a CSV file with one row per segment (start time, end time, text, confidence), ready to open in a spreadsheet.
The JSON file includes the following fields for each segment:
| Column | Description |
|---|---|
start_ms | Segment start time in milliseconds from the recording start |
end_ms | Segment end time in milliseconds from the recording start |
text | Transcribed text for the segment |
Accuracy considerations
- Accuracy is highest with clear audio, a single speaker, and minimal background noise.
- Selecting the correct language is important; mismatched language settings will reduce accuracy significantly.
- Recording conditions (microphone quality, room echo, participant distance from the microphone) affect the result.
- The transcript is automatic and may contain errors, particularly for technical vocabulary, names, or accented speech. Always review the transcript before using it for analysis.
Troubleshooting
The analysis shows "Failed" The most common causes are a recording with no audio track or a server-side issue during processing. You can restart the analysis from the Analysis tab. If it fails again, check whether the recording contains audible speech.
The transcript is blank or very short This may happen if the participant was silent, the microphone was muted, or the audio track was not captured. Review the recording to confirm audio is present.
Timestamps seem off Transcription timestamps are aligned to the recording start. If the recording itself has synchronisation issues, the timestamps will reflect that. Check the sync quality indicator on the participation page.