Datasets:
questions on dataset preparation using pyannote.audio
#5
by
rjrobben
- opened
Can I ask how do you recognise the speaker using pyannote.audio? It works when there is only one short audio file(say 20 mins), speakers are identified with index, SPEAKER000, SPEAKER001 and so on.
But when having lots of audio files, each with different sets of speakers, how do you identify the speaker and assign a unique id by pyannote.audio?
One naive way is to combine all of them to a single file, when the recordings are long (I tried 20 hours wav), the diarization process requires so much ram and the system crashes.
Just wonder how do you resolve this problem in your dataset preparation process, do we have to rely on the embeddings and run ml algorithms? Would be grateful if you can provide some insights. Thanks!