Diarization enhancements + dependencies update #157

chainyo · 2023-07-18T16:04:12Z

This PR:

Improve transcription & diarization mapping process to use transcription segments and word timestamps for mapping.
Bump important dependencies versions to benefit from recent updates in faster-whisper and NeMo.
Update pydantic code to use v2 benefits and last fastapi releases.
Add float32 as a valid possibility for the model compute type.

chainyo added 7 commits July 17, 2023 15:56

first attempt to fix speaker attribution

ec56949

add threshhold to diarization strategy

114bcd6

bump deps versions

9abefcc

clean new implementation

a4eea63

update for pydanticv2

a9419de

fix models for pydanticv2

12a7984

bump faster-whisper version

48dde22

chainyo linked an issue Jul 18, 2023 that may be closed by this pull request

Update post-processing strategy #153

Closed

aleksandr-smechov approved these changes Jul 18, 2023

View reviewed changes

chainyo added 5 commits July 19, 2023 09:02

update models for pydanticv2

19adcfd

fix tests

c90dc64

update nemo version

48fe805

fix linter

944a7ed

fix test typo

d5568b4

aleksandr-smechov approved these changes Jul 19, 2023

View reviewed changes

linter

7a72c4a

aleksandr-smechov approved these changes Jul 19, 2023

View reviewed changes

chainyo merged commit b2b261c into main Jul 19, 2023
3 checks passed

chainyo deleted the 153-update-post-processing-strategy branch July 19, 2023 13:17

Provide feedback