m-bain/whisperX

This system is a pre-existing audio transcription tool, likely based on WhisperX, that processes audio streams into text. Recent development has focused on stabilizing the environment through Windows AMD64 support and improved NLTK error handling, while introducing a new interleaved context feature to maintain coherence across multiple concurrent audio streams. The codebase is currently in a maintenance and incremental feature-addition phase, with no complex microservices or bounded contexts evident in its structure.

Narrated so far: 2026-06-03 – 2026-09-06; history before 2026-06-03 not narrated yet.

• Updated dependency markers to support Windows AMD64 architecture — The pyproject.toml dependency configuration has been updated to correctly detect and select packages for Windows systems running on AMD64 processors. Previously, the platform markers only checked for 'x86\_64', which excluded Windows AMD64 machines from receiving the appropriate PyTorch and Torchaudio builds. The new markers explicitly include 'AMD64' alongside 'x86\_64' for non-Darwin systems, ensuring that users on Windows AMD64 hardware receive the correct CPU or GPU variants of these libraries. ((dependencies))

1 change.

The huggingface-hub dependency lower bound has been tightened to version 0.28.1, replacing the previous upper-bound constraint. The package version has also been incremented to 3.8.7rc1.

1 change.

When the required NLTK 'punkt\_tab' data is missing, the system now attempts to download it and provides a clear, actionable error message if the download fails, including instructions on how to install it manually. Previously, a silent failure or generic exception could occur if the download did not succeed, making troubleshooting difficult for users with network issues or restricted environments.

1 change.

Users can now enable the new --interleaved\_context CLI flag to improve transcription continuity across multiple concurrent audio streams. When active, WhisperX passes the final tokens from previous batches as context to subsequent batches, allowing the model to maintain coherence across stream boundaries. This feature is gated by the batch size (requiring batch\_size \> 1) and includes warnings if the batch size is too large relative to the number of audio segments, which would render the context rolling ineffective.

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

• First survey — no prior run to compare against. CAI 61.

• Code Health 87
• Architecture 100
• Maturity 55
• Readiness 52
• Security 74

• 300 commits — 239 feature/other, 61 fixes

• (root) — 92 commits
• (repo) — 69 commits
• whisperx/alignment.py — 46 commits
• whisperx/asr.py — 30 commits
• .github/workflows — 13 commits
• whisperx/vads — 10 commits
• whisperx/transcribe.py — 8 commits
• whisperx/utils.py — 8 commits
• whisperx/main.py — 6 commits
• whisperx/diarize.py — 6 commits
• tests/test_word_timestamp_interpolation.py — 3 commits
• build/lib — 2 commits
• whisperx/SubtitlesProcessor.py — 2 commits
• whisperx/vad.py — 2 commits
• models/pytorch_model.bin — 1 commit
• whisperx/init.py — 1 commit
• whisperx/assets — 1 commit

• fix: fix: pin huggingface-hub<1.0.0 for pyannote-audio compatibility (#1327)
• fix: FIX warnings for word options
• fix: Fix Windows CUDA detection: include AMD64 in platform markers (#1357)
• fix: Fix link in README.md
• fix: Fix: Allow vad options to be configurable by correctly passing down to FasterWhisperPipeline.
• fix: Fix: Ensure integer tensor indexing in get_wildcard_emission()
• fix: Fixes --model_dir path
• fix: Merge pull request #1342 from 1carlito/bugs
• fix: Merge pull request #1343 from m-bain/fix-type-hint-decode-batch
• fix: Merge pull request #1356 from m-bain/revert-1355-batch_wrap
• fix: Merge pull request #438 from invisprints/fix-speaker-missing
• fix: Merge pull request #473 from sorgfresser/fix-faster-whisper-threads
• fix: Merge pull request #507 from compasspathways/fix/pass-vad-options
• fix: Merge pull request #554 from sorgfresser/fix-binarize-unbound
• fix: Merge pull request #716 from cococig/fix/faster-whisper-from-pypi
• fix: Revert "Batch wrap"
• fix: Revert "feat: add Basque alignment model (#1074)" (#1077)
• fix: [BugFix] The variable I removed was not being used anyhwere.
• fix: [BugFix] Type hint fix in decode_batch List[str] not str:
• fix: [fix] Batch context is updated each time. It works with the initial prompt added.
• …and 280 more

Written by watchdog.canine.dev from the codebase's own history, inside the signed delivery this page is composed from.

m-bain/whisperX was measured the same way every project in this corpus was: the same rubric, at a pinned commit, with the result published in full. Point a surveyor at a repository you know and see whether you agree with it.