This is an old revision of the document!
Find and Review the Results
Open the output directory in the portal file browser. Whisper writes five files for each recording, all sharing the recording's name:
| Extension | Intended use |
|---|---|
.txt | Plain transcript text, without timings |
.srt | Subtitles with timestamps, for media players and editors |
.vtt | Web subtitles, also with timestamps |
.json | Structured output for further processing |
.tsv | A table of start, end and text, in milliseconds |
A recording named interview.mp3 therefore produces interview.txt, interview.srt, and so on. If you need the transcript as a document, start from the .txt file; if you need subtitles, use .srt or .vtt.
The Job Log
The job log is written next to the job script, in the Script Location directory from the form, and not in the output directory. For the default location that means:
/scratch/<your account>/whisper/slurm-<JobID>.out
It records what Whisper did, including the model it loaded and any errors. Read it when the output is missing or looks wrong, and quote it when you contact support.
Check the Transcript
Before relying on a transcript:
- Read the whole transcript against the recording.
- Verify names, dates, numbers, quotations, and specialist terms.
- Check speaker labels if diarization was enabled.
- Check subtitle timing if you plan to publish subtitles.
- Save the reviewed result somewhere intended for long-term storage.
You do not have to finish this review on Hábrók. Download the results to your own work environment first; a quick check that the transcription succeeded is enough to decide that the job is done, and the detailed reading can follow later. Once you have the files, remove them from Hábrók as described in Data management safety measures.