How improve Watson Speech to Text speaker label

1
votes

I'm using Watson Speech to Text on my projects, and I have questions.

  • How can I improve the quality of speaker labeling?

  • Does acoustic custom model effect this improvement?

    Please give me some advice.

1
ibm-cloudibm-watsonspeech-to-text
**My use case is like a small meeting for 4~10 speakers , but in each recording, the speakers always change (over 100 people in total). I only need to identify 4~10 speakers in each meeting, and not need to identify all (over 100) speakers in the long term – Yuki Ushioda

1 Answers

0
votes

Acoustic model customization will only have a moderate impact in the quality of the speaker labels. It should get better, but do not expect drastic improvements.

What is your use case for speaker labels? how many speakers do you have in your recording?