with PyAnnote, Nvidia NeMo and Pipecat Smart Turn
Apologies, I pressed send on this "Turn Detection and Diarization" and then realised I've a very slow internet connection and the YouTube video is not live yet. It should be by 6 pm Irish time today Monday 17th March 2025. Apologies
I am using also different prompts for diarisation with score to give a name to speakers..
Interesting, can you explain a bit more?
After diarization my goal is to identify and give a name to all speakers
so i use llm to try to identify them
easy one is if a speaker present himself, ....
Ah yes, nice.
Another approach is to match speaker embeddings with the diarised audio.
But this requires having a sample for each speaker - and their name - in advance
Yes and in france we have rgpd..
At the end the admin can also add them or an ai can ask to every unknow speaker to identifu themself outputing a sample of audio for each
Apologies, I pressed send on this "Turn Detection and Diarization" and then realised I've a very slow internet connection and the YouTube video is not live yet. It should be by 6 pm Irish time today Monday 17th March 2025. Apologies
I am using also different prompts for diarisation with score to give a name to speakers..
Interesting, can you explain a bit more?
After diarization my goal is to identify and give a name to all speakers
so i use llm to try to identify them
easy one is if a speaker present himself, ....
Ah yes, nice.
Another approach is to match speaker embeddings with the diarised audio.
But this requires having a sample for each speaker - and their name - in advance
Yes and in france we have rgpd..
At the end the admin can also add them or an ai can ask to every unknow speaker to identifu themself outputing a sample of audio for each