Skip to content

parakeet multitalker with diarization - #122

Merged
cjpais merged 3 commits into
mainfrom
sortformer-multitalker
Aug 7, 2026
Merged

parakeet multitalker with diarization#122
cjpais merged 3 commits into
mainfrom
sortformer-multitalker

Conversation

@cjpais

@cjpais cjpais commented Jul 30, 2026

Copy link
Copy Markdown
Contributor

No description provided.

@cjpais
cjpais merged commit 62e7769 into main Aug 7, 2026
25 checks passed
cjpais added a commit that referenced this pull request Aug 7, 2026
…p docs

Review follow-ups to #122 that missed the merge:

- e2e smoke: select each supervision mode explicitly (the "masked" leg
  previously unset TRANSCRIBE_MULTITALKER_MODE and ran the kernel
  default twice; masked mode was never exercised)
- run_cpp_multitalker.py: forward --mode verbatim (masked previously
  fell through to the kernel default); temp files under repo tmp/
- compose-multitalker-bundle.py: pin the diarizer to validated variants
  (the runtime hardcodes the v2.1 operating point) and enforce
  ASR/diarizer subsampling agreement (the supervision hand-off assumes
  matching 80 ms frame cadences)
- multitalker.cpp: correct the memory-shape comments (O(1) per-speaker
  working state, O(T) call memory), the stale masked cpWER figure, and
  the mode-selection contract comment
- docs: bundle download table + CLI example, capability-table rows
  flipped to PASS with cpWER evidence, intake retro
- HF card: diarization tags/summary, diarize capability flag, and a
  metrics: block (cpwer_ami_ihm_test) with generate.py support
- gen_multitalker_oracle_audio.py: uv script header, portable sha256

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant