Azure batch transcription diarization silently ignored on stereo audio

TL;DR

Batch transcription with the DiarizationEnabled property set produces transcription JSON with no Speaker property at all. The silent ignore was a bug on the service side; a fix was being deployed to return a clear error message instead of doing nothing.

Steps

1. Batch transcription with the DiarizationEnabled property set produces

transcription JSON with no Speaker property at all. The flag looks ignored. Multiple audio files, including ones with clearly distinct male and female voices, all come back with no speaker labels.

2. From the service logs: the input file was stereo, and for stereo files

diarization is not possible, so the service silently ignored the diarization flag. The silent ignore was a bug on the service side; a fix was being deployed to return a clear error message instead of doing nothing. But the underlying constraint remains.

3. Downmix your audio to mono before submitting the transcription request. The

Speaker property then shows up in the results.

4. Check your input: ffprobe will tell you the channel count.

5. If stereo, downmix: ffmpeg -i input.wav -ac 1 mono.wav (or do it in your audio pipeline).

6. Resubmit the mono file with DiarizationEnabled set.

When to use

You are seeing this: Multiple audio files, including ones with clearly distinct male and female voices, all come back with no speaker labels. Use this skill when you run into "Azure batch transcription diarization silently ignored on stereo audio".

When not to use

If your error message or symptom does not match what is described above, this is probably not your fix. Search for your exact error text instead of forcing this one to fit.

Versions

No specific versions are mentioned in the source material, so treat the fix as generally applicable and check the examples against whatever you have installed.

Why this happens

From the service logs: the input file was stereo, and for stereo files diarization is not possible, so the service silently ignored the diarization flag.

A user asked why Azure batch transcription with DiarizationEnabled returned no Speaker property on any audio file, including files with clearly distinct voices. The verified resolution: from the service logs, the input file was stereo and diarization is not possible on stereo files, so the flag was silently ignored (a bug being fixed to return a clear error); downmixing the audio to mono before submitting makes the Speaker property show up. Thread: https://vectle.com/threads/thr_ou0QQEdNI03jpdj87BzUXA