Commit graph

  • b102e12943 M5 benchmark figures: WER vs RTF scatter, 0.6B+1.7B MLX results main Quentin Fuxa 2026-03-15 15:00:00 +0100
  • 7aa3b764bd MLX benchmark: 1.7B SimulStreaming on M5 (WER 4.07%, RTF 0.944) Quentin Fuxa 2026-03-15 14:00:00 +0100
  • a422e604ae MLX benchmark: 0.6B SimulStreaming on M5 MacBook (WER 3.30%, RTF 0.263) Quentin Fuxa 2026-03-15 13:00:00 +0100
  • e14b913807 Merge branch 'benchmarks-h100' Quentin Fuxa 2026-03-15 12:00:00 +0100
  • 47d4cbeecc reorganize benchmarks: move H100 results to benchmarks/h100/ Quentin Fuxa 2026-03-15 23:59:00 +0100
  • 3b7a2fcc87 Add Qwen3-ASR MLX SimulStreaming backend Quentin Fuxa 2026-03-15 11:00:00 +0100
  • f75dfb386d final benchmark: Voxtral vLLM realtime streaming Quentin Fuxa 2026-03-15 23:59:00 +0100
  • 276ba84d02 update figures with Voxtral vLLM results Quentin Fuxa 2026-03-15 23:55:00 +0100
  • 36b3885cf2 add Voxtral 4B to benchmark figures Quentin Fuxa 2026-03-15 23:30:00 +0100
  • a29e799ba5 update H100 benchmark figures with ACL6060 results Quentin Fuxa 2026-03-15 22:30:00 +0100
  • 22325ba326 tune simul-kv: 2s inference interval, configurable min_new_seconds Quentin Fuxa 2026-03-15 21:30:00 +0100
  • a540a5fd10 fix simul-kv audio trim bug, add 1.7B v2 alignment heads Quentin Fuxa 2026-03-15 20:45:00 +0100
  • 7b08ea74ab add H100 benchmark figures Quentin Fuxa 2026-03-15 19:15:00 +0100
  • b69eaf82be qwen3 simul+kv: optimized streaming with kv cache reuse Quentin Fuxa 2026-03-15 18:30:00 +0100
  • ed503be140 qwen Quentin Fuxa 2026-01-02 23:52:00 +0100
  • a6a85431f6 update benchmark with qwen3 which reuses kv cache Quentin Fuxa 2026-03-15 22:32:01 +0100
  • dd48997674 qwen3: reuse encoder kv cache Quentin Fuxa 2026-03-15 22:31:39 +0100
  • f24481dc29 update archi Quentin Fuxa 2026-03-15 11:36:45 +0100
  • ed76f40ee5 Merge branch 'main' of https://github.com/QuentinFuxa/WhisperLiveKit Quentin Fuxa 2026-03-15 11:16:38 +0100
  • 5330b3fac5 update benchmark part Quentin Fuxa 2026-03-15 11:16:26 +0100
  • 0c73a73aa3 update benchmark results and procedure Quentin Fuxa 2026-03-15 11:16:15 +0100
  • 2d6bc4f572
    Add '*.c' to .dockerignore Quentin Fuxa 2026-03-14 00:18:10 +0100
  • dfd5bf417c voxtral mlx : improved chunking Quentin Fuxa 2026-03-14 00:13:29 +0100
  • 9d8db7ab38 add qwen3 simul in tests Quentin Fuxa 2026-03-14 00:13:09 +0100
  • fa15115163 qwen3 alignment heads Quentin Fuxa 2026-03-14 00:12:50 +0100
  • 8dc7b77071 Bump version to 0.2.20 Quentin Fuxa 2026-03-08 16:02:00 +0100
  • 10d85ff65f Update docs, CI, and architecture diagram Quentin Fuxa 2026-03-08 15:14:00 +0100
  • e7e3441ca4 Add Qwen3 ASR backend Quentin Fuxa 2026-03-07 11:48:00 +0100
  • 9abe26a996 Add CLI with serve, transcribe, listen, pull, diagnose Quentin Fuxa 2026-03-01 13:37:00 +0100
  • c8e7c216ed Replace mock tests with real pipeline tests Quentin Fuxa 2026-02-28 10:05:00 +0100
  • 586540ae36 Add test harness and test client Quentin Fuxa 2026-02-22 16:19:00 +0100
  • cd8df8e1aa Update package setup and exports Quentin Fuxa 2026-02-21 11:33:00 +0100
  • e30f9a2573 Improve diarization backends Quentin Fuxa 2026-02-15 14:55:00 +0100
  • 32de7b1276 Fix frontend buffer rendering for slow backends Quentin Fuxa 2026-02-14 09:28:00 +0100
  • 9ac7c26a0b Add OpenAI REST API and Deepgram WebSocket Quentin Fuxa 2026-02-08 15:42:00 +0100
  • c0e2600993 Add snapshot-then-diff WebSocket protocol Quentin Fuxa 2026-02-07 10:17:00 +0100
  • e0db3a98f9 Add per-session language proxy Quentin Fuxa 2026-02-01 17:03:00 +0100
  • 2fe34427ef Fix voxtral streaming drain and silence flush Quentin Fuxa 2026-01-31 11:12:00 +0100
  • d58365421f Refactor audio processor async pipeline Quentin Fuxa 2026-01-25 13:48:00 +0100
  • a282cbe75f Improve tokens alignment and silence handling Quentin Fuxa 2026-01-24 10:55:00 +0100
  • 6e85c16614 Refactor TranscriptionEngine singleton Quentin Fuxa 2026-01-18 15:27:00 +0100
  • e1823dd99c Improve online ASR processor Quentin Fuxa 2026-01-17 09:35:00 +0100
  • e144abbbc7 Refactor timed objects and data structures Quentin Fuxa 2026-01-11 16:08:00 +0100
  • 83362c89c4 Clean up config and model paths Quentin Fuxa 2026-01-10 11:42:00 +0100
  • 74c4dc791d Lint scripts and tests Quentin Fuxa 2026-01-04 14:15:00 +0100
  • cf6c49f502 Ruff lint cleanup Quentin Fuxa 2026-01-03 10:23:00 +0100
  • 451535d48f Fix ctranslate2 encoder conversion (#345) and memory leak in TokensAlignment (#344) Quentin Fuxa 2026-03-10 22:37:00 +0100
  • 8bc0937c46
    Update README section on powered research Quentin Fuxa 2026-03-06 18:46:07 +0100
  • 929cf7a26b add link to AlignAtt interactive playground Quentin Fuxa 2026-03-06 18:43:25 +0100
  • abfaf06203 Merge branch 'main' of https://github.com/QuentinFuxa/WhisperLiveKit Quentin Fuxa 2026-03-04 18:17:23 +0100
  • d1fe932241 Apply DRY method v0 - to try to catch and resolve infinite loops such as in #338 Quentin Fuxa 2026-03-03 22:52:00 +0100
  • c112ceffb6
    Merge pull request #342 from mnicnc404/fix/whisper-tokenizer-index-error Quentin Fuxa 2026-03-02 20:36:58 +0100
  • 4917406e06
    Merge pull request #341 from AymurAI/feat/uv-deps-resolution Quentin Fuxa 2026-03-02 20:34:49 +0100
  • b63f54e838 fix(whisper/tokenizer): prevent IndexError from crashing multilingual streams Chingning Chen 2026-03-02 15:31:43 +0800
  • c56a53fbf4 deps(mlx-groups): add optional dependencies for Apple Silicon MLX backends jedzill4 2026-03-01 20:05:52 -0300
  • 66e58624b9 disable MLXAlignAtt which fails on special characters Quentin Fuxa 2026-03-01 11:52:00 +0100
  • 9366e067f9 deps(pyproject): add torch and torchaudio to main dependencies jedzill4 2026-02-27 19:19:18 -0300
  • 866c25670c deps(docker): change CUDA base image to runtime version jedzill4 2026-02-27 19:16:29 -0300
  • 2553ef283e deps(docker): fix dependency group for cu129 image jedzill4 2026-02-25 21:49:08 -0300
  • 73e7fafc48 feat(tests): python matrix support test jedzill4 2026-02-25 21:35:41 -0300
  • bbcebcb1fe deps(sortformer): adjust nemo-toolkit version constraints jedzill4 2026-02-25 21:33:00 -0300
  • 4bb58dc7aa deps(diart): improve diart dependency tree. rename gpu-cu129 dependency group to cu129 jedzill4 2026-02-25 20:27:26 -0300
  • 27ca028479 ci(github): add GitHub Actions workflows for Docker image publishing and support matrix jedzill4 2026-02-25 14:27:51 -0300
  • d24805cc18 🚀 chore (docker): update docker images improving caching and using uv as python package manager jedzill4 2026-02-25 14:22:43 -0300
  • 994ce21365 📌 chore(deps): pin dependences to python 3.11 to 3.13 due dependency resolution matrix jedzill4 2026-02-25 14:21:19 -0300
  • 132823dc09 deps: improve deps dependency resolution (wip) jedzill4 2026-02-24 20:15:53 -0300
  • d6d8c2635f chore: use uv as python project manager to improve dependency resolution jedzill4 2026-02-23 22:16:32 -0300
  • 8fedeb9fed
    Merge pull request #340 from QuentinFuxa/voxtral_tests Quentin Fuxa 2026-02-23 10:37:40 +0100
  • b1fc23807a docs: add benchmark collaboration call, voxtral in powered-by section Quentin Fuxa 2026-02-23 10:37:22 +0100
  • 10c4e5f730 docs: add speed vs accuracy scatter plot to benchmark and README Quentin Fuxa 2026-02-23 10:27:53 +0100
  • c76b2ef2c6 docs: rewrite benchmark with base/small comparison, proper French results Quentin Fuxa 2026-02-23 10:16:34 +0100
  • 4b2377c243 fix: correct false auto-detect claim, median bug, RTF inflation Quentin Fuxa 2026-02-22 23:38:04 +0100
  • a4da246ea5 feat: add voxtral-mlx native backend for Apple Silicon Quentin Fuxa 2026-02-22 23:28:10 +0100
  • 9b2c3ee844 docs: update README with voxtral backend, benchmarks, testing sections Quentin Fuxa 2026-02-22 23:27:57 +0100
  • 83d0fa3fac feat: benchmark suite with WER, timestamp accuracy, cross-backend comparison Quentin Fuxa 2026-02-22 23:27:50 +0100
  • 5a12c627b4 feat: add 99-test unit test suite with zero model dependencies Quentin Fuxa 2026-02-22 23:27:40 +0100
  • f5eee67b11 fix: silence double-counting bug, add metrics module and runtime instrumentation Quentin Fuxa 2026-02-22 23:27:12 +0100
  • 4a6868e3e1 correct processor attributes mixtral Quentin Fuxa 2026-02-22 21:13:21 +0100
  • 3c15246fc0 mixstral hf v0 Quentin Fuxa 2026-02-20 20:46:37 +0100
  • d337248fda feat: add healthcheck to Dockerfiles (#228) Quentin Fuxa 2026-02-19 22:18:00 +0100
  • b8d9d7d289 fix: handle numpy object_ dtype from ctranslate2 encoder (#337) Quentin Fuxa 2026-02-19 22:18:00 +0100
  • 4c7706e2cf fix: use vac_chunk_size for audio processing interval when VAC is enabled (#334) Quentin Fuxa 2026-02-19 22:18:00 +0100
  • 7f3a3df620 simulstreaming mlx & torch dedup of common base Quentin Fuxa 2025-02-15 23:52:00 +0100
  • e7e82f7c19 bump to 0.2.18 Quentin Fuxa 2026-02-11 22:10:00 +0100
  • 8c799fa4d1 fix simulstreaming vram leak: cap cross-attn accumulation + token budget Quentin Fuxa 2026-02-11 22:10:00 +0100
  • 8923337380 fix --direct-english-translation not setting task=translate for localagreement backends Quentin Fuxa 2026-02-11 22:10:00 +0100
  • aded1649ae fix model_cache_dir + direct_english_translation task in simulstreaming Quentin Fuxa 2026-02-11 22:10:00 +0100
  • 3b535e857a fix NoneType concatenation in add_translation Quentin Fuxa 2026-02-11 22:10:00 +0100
  • d649250b9a fix Segment classmethod call + isinstance type narrowing Quentin Fuxa 2026-02-11 22:10:00 +0100
  • 7735478286 add insert_audio_chunk to DiartDiarization Quentin Fuxa 2026-02-11 22:10:00 +0100
  • b9e72d2b9a add probability field to ASRToken Quentin Fuxa 2026-02-11 22:10:00 +0100
  • e5b01033af add json normalizers for english language in build Quentin Fuxa 2026-01-16 10:47:46 +0100
  • 6ae545bcb1 bump to 0.2.17.post1 Quentin Fuxa 2026-01-16 10:43:52 +0100
  • 04980d3f5e Merge branch 'main' of https://github.com/QuentinFuxa/WhisperLiveKit Quentin Fuxa 2026-01-16 10:38:29 +0100
  • 79a705c969 fixes #323 Quentin Fuxa 2026-01-16 10:38:07 +0100
  • 34e4abd455
    Merge pull request #322 from eschmidbauer/fix/thread-safety-issues Quentin Fuxa 2026-01-09 19:23:35 +0100
  • d59ddbaeae Fix critical thread safety issues Emmanuel Schmidbauer 2026-01-09 11:23:19 -0500
  • 4dd66e7766
    Merge pull request #317 from jantonj/fix-bug-diarization-lag Quentin Fuxa 2025-12-19 17:43:07 +0100
  • 3db5d81a20 update diarization lag after stream analysed Anton Jacobson 2025-12-18 14:13:28 +0100
  • b67ddea494 bump to 0.2.17 Quentin Fuxa 2025-12-08 23:52:00 +0100