Skip to content

Add Whisper word-timing alignment and timestamps skill - #37

Merged
andyhtran merged 2 commits into
mainfrom
ft-cli
Jul 8, 2026
Merged

Add Whisper word-timing alignment and timestamps skill#37
andyhtran merged 2 commits into
mainfrom
ft-cli

Conversation

@andyhtran

Copy link
Copy Markdown
Owner

Summary

  • Add --align <none|whisper-token|whisper-dtw> flag for Whisper word-level timing with DTW/flash-attention mutual exclusion and VAD bypass when token timestamps are active
  • Extract token timings from whisper.cpp and merge into word-level outputs via WhisperWordTimingBuilder
  • Split agent skill into separate core and timestamps guides; refactor core skill to be concise with cross-reference to timestamps skill
  • Rework CLI help tiers, add multi-target skill install, and Codex support
  • Allow --align none as a no-op with Parakeet (only error on non-none alignment modes)

Test plan

  • swift build succeeds
  • swift test --quiet passes
  • miniwhispercli transcribe <audio> --model whisper --align whisper-token --format json produces word timings
  • miniwhispercli transcribe <audio> --model whisper --align whisper-dtw --format json produces DTW word timings
  • --align none with Parakeet proceeds without error
  • --align whisper-token with Parakeet returns usage error
  • miniwhispercli skills get timestamps prints the timestamps guide
  • miniwhispercli skills list --json includes the timestamps skill

🤖 Generated with Claude Code

andyhtran and others added 2 commits July 7, 2026 20:25
Introduce bare/standard/full help levels with --full flag for the complete
command reference. Add --target/--codex/--skill-dir flags to skill commands
for Codex and custom skill directory support. Replace document-skills path
with per-agent skill directories, add dry-run preview for skill install,
and restructure skills list/get with JSON output and runtime skill registry.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
…ore skill

Add --align flag (whisper-token, whisper-dtw) for Whisper word-level timing
with DTW/flash-attention mutual exclusion and VAD bypass when token timestamps
are active. Extract token timings from whisper.cpp and merge into word-level
outputs. Split agent skill into core and timestamps guides. Allow --align none
as a no-op with Parakeet.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
@andyhtran
andyhtran merged commit 261b488 into main Jul 8, 2026
1 check passed
@andyhtran
andyhtran deleted the ft-cli branch July 8, 2026 02:36
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