transcribe: add speaker diarization test #12410
Conversation
34b15db to
71eba5b
Compare
There was a problem hiding this comment.
I doubt this test will be useful. Without the actual models used by AWS Transcribe, we will not be able to reproduce the same timing and other numerical values. It's the same reason we can't AWS validate transcription tests. That said, it doesn't hurt to have this in as a reference when diarisation is being added.
|
Thanks for the review 🙌 I have slightly different viewpoint here, although I agree that we won't be able to reproduce the same timing and other numerical values without the actual models being used by AWS Transcribe, but in my opinion this test could be really helpful in kicking off the speaker diarization implementation in future, where we could atleast know the currently missing attributes and functionality considering the logged AWS responses, also making it easier to understand and implement the feature of atleast having This could be later modified according to the then implementation and skip the snapshot of certain functionality. |
Motivation
Add a test to Localstack Transcribe with an enabled speaker partitioning/diarization. Speaker partitioning labels the speech from individual speakers in the media file.
This can be achieved by enabling
ShowSpeakerLabelsin the transcription settings and specifying theMaxSpeakerLabelswhich is the maximum number of speakers we want to partition in the media file.Changes
This PR:
tests/aws/files/multi-speaker.wavwith 2 speakers.