Code

Add speaker-aware transcription#147

Code

Draft

juanmc2005 wants to merge 26 commits intodevelopjuanmc2005/diart:developfrom feat/spk-asrjuanmc2005/diart:feat/spk-asrCopy head branch name to clipboard

+2,157-775

Commits

Commits on Apr 19, 2023

• New feature: streaming voice activity detection. Pipeline name changes

  juanmc2005

  committed
  bca2873

  Copy full SHA for bca2873

• Merge branch 'develop' of github.com:juanmc2005/OnlineDiarization into feat/vad

  juanmc2005

  committed
  5e44ad4

  Copy full SHA for 5e44ad4

• Update link in setup.cfg

  juanmc2005

  committed
  7447061

  Copy full SHA for 7447061

• Update code snippets in README

  juanmc2005

  committed
  4985394

  Copy full SHA for 4985394

• Add minor README modifications

  juanmc2005

  committed
  540ad0a

  Copy full SHA for 540ad0a

Commits on Apr 21, 2023

• Initial ASR implementation. Broken stuff

  juanmc2005

  committed
  8cc9925

  Copy full SHA for 8cc9925

• First working transcription pipeline. Using diarization is possible but a bit quirky

  juanmc2005

  committed
  1ae4934

  Copy full SHA for 1ae4934

• Reduce Whisper VRAM footprint (around 400Mb). Add fp16 option

  juanmc2005

  committed
  d8d7342

  Copy full SHA for d8d7342

• Change whisper input type based on fp16 parameter

  juanmc2005

  committed
  2cfc35d

  Copy full SHA for 2cfc35d

Commits on Apr 22, 2023

• Implement batched inference for whisper. Re-implement decoding.

  juanmc2005

  committed
  a40112c

  Copy full SHA for a40112c

• Minor changes in transcription arguments

  juanmc2005

  committed
  e8196a7

  Copy full SHA for e8196a7

Commits on Apr 23, 2023

• Greatly improve transcription pipeline by adding optional VAD

  juanmc2005

  committed
  07dd9ae

  Copy full SHA for 07dd9ae

• Move pipelines to diart.pipelines. Add torchmetrics as a dependency

  juanmc2005

  committed
  0bf2522

  Copy full SHA for 0bf2522

• Add websocket compatibility to transcription pipeline

  juanmc2005

  committed
  42fe5f7

  Copy full SHA for 42fe5f7

• Transcription pipeline is now fully compatible with diart.stream

  juanmc2005

  committed
  49616e5

  Copy full SHA for 49616e5

Commits on Apr 24, 2023

• Make transcription pipeline compatible with diart.benchmark and diart.tune. Fix major bug in Optimizer

  juanmc2005

  committed
  babf49d

  Copy full SHA for babf49d

• Rename base pipeline and config objects

  juanmc2005

  committed
  6609e3c

  Copy full SHA for 6609e3c

• Merge changes from branch feat/vad

  juanmc2005

  committed
  4c1aeba

  Copy full SHA for 4c1aeba

• New feature: streaming voice activity detection. Pipeline name changes

  juanmc2005

  committed
  d19b044

  Copy full SHA for d19b044

• Update link in setup.cfg

  juanmc2005

  committed
  6caa4a4

  Copy full SHA for 6caa4a4

• Update code snippets in README

  juanmc2005

  committed
  0993fe8

  Copy full SHA for 0993fe8

• Add minor README modifications

  juanmc2005

  committed
  95d4fae

  Copy full SHA for 95d4fae

• Rename base pipeline and config objects

  juanmc2005

  committed
  569c68f

  Copy full SHA for 569c68f

• Update branch with develop

  juanmc2005

  committed
  eed864f

  Copy full SHA for eed864f

• Add initial implementation of SpeakerAwareTranscription

  juanmc2005

  committed
  a16bb5c

  Copy full SHA for a16bb5c

Commits on Apr 26, 2023

• Refactor SpeakerAwareTranscription

  juanmc2005

  committed
  c7bbcc4

  Copy full SHA for c7bbcc4