v3.4.2

What's Changed

• Fix: Ensure integer tensor indexing in get_wildcard_emission() to avoid IndexError by @HowardWhile in #1146

New Contributors

• @HowardWhile made their first contribution in #1146

Full Changelog: v3.4.1...v3.4.2

v3.4.1

What's Changed

• fix: speaker embedding bug by @Barabazs in #1178

Full Changelog: v3.4.0...v3.4.1

v3.4.0

What's Changed

• chore: add lockfile check step to CI workflows by @Barabazs in #1130
• docs: add common issue section for libcudnn dependencies in README by @Barabazs in #1161
• feat: diarization model env config by @bgdnvk in #1101
• docs: add missing torch import to Python usage example in README by @hammerill in #1168
• feat: enhance diarization with optional output of speaker embeddings by @eek in #1085

New Contributors

• @bgdnvk made their first contribution in #1101
• @hammerill made their first contribution in #1168
• @eek made their first contribution in #1085

Full Changelog: v3.3.4...v3.4.0

v3.3.4

What's Changed

• feat: improve CLI loading speed by lazy loading public API by @Barabazs in #1128

Full Changelog: v3.3.3...v3.3.4

v3.3.3

What's Changed

• Silero VAD support by @3manifold in #888
• refactor: add more types by @Barabazs in #996
• fix vad_method is none by @winking324 in #995
• support timestamp for numbers. by @bfs18 in #986
• Update links to language models in README by @MJochim in #757
• chore: remove deprecated VAD_SEGMENTATION_URL by @Barabazs in #1003
• chore: handle empty segments_list case in silero by @tan90xx in #1005
• chore: fix variable naming inconsistency from segments to segments_list by @tan90xx in #1006
• feat: add Latvian align model by @slikts in #1017
• Add models_cache_only param by @philmcmahon in #1024
• Added Phoneme-Based ASR Model for Tagalog by @mtfugin in #1067
• Basque alignment model by @xezpeleta in #1074
• Revert "feat: add Basque alignment model (#1074)" by @Barabazs in #1077
• feat: use uv for package/project management by @Barabazs in #1002
• feat(hotwords): Pass hotwords option to faster-whisper by @jademlc in #1073
• refactor: update import statements to use explicit module paths by @Barabazs in #1091
• docs: update installation instructions by @Barabazs in #1092
• fix: update setuptools configuration to include package discovery by @Barabazs in #1093
• Remove duplicated item by @yccheok in #1109
• fix: resolve dependency issues by @Barabazs in #1126

New Contributors

• @3manifold made their first contribution in #888
• @winking324 made their first contribution in #995
• @bfs18 made their first contribution in #986
• @MJochim made their first contribution in #757
• @tan90xx made their first contribution in #1005
• @slikts made their first contribution in #1017
• @philmcmahon made their first contribution in #1024
• @mtfugin made their first contribution in #1067
• @xezpeleta made their first contribution in #1074
• @jademlc made their first contribution in #1073
• @yccheok made their first contribution in #1109

Full Changelog: v3.3.1...v3.3.3

v3.3.2

What's Changed

• chore: update ctranslate2 version requirement to >=4.5.0

This patch release should resolve CUDNN related issues.

Full Changelog: v3.3.1...v3.3.2

v3.3.1

What's Changed

• refactor: add type hints and fix import statement by @Barabazs in #975
• feat: include speaker information in WriteTXT when diarizing by @Barabazs in #976
• Bug Fix: Suppress Numerals dataclasses replace method by @jmt0221 in #981

New Contributors

• @jmt0221 made their first contribution in #981

Full Changelog: v3.3.0...v3.3.1

v3.3.0

What's Changed

• Update faster-whisper to 1.0.2 to enable model distil-large-v3 by @moritzbrantner in #814
• latest faster-whisper support added by @Hasan-Naseer in #875
• Working version with pyannote:3.3.2 and faster-whisper:1.1.0 by @ibombonato in #936
• Add ultization to verbose flag by @H4CK3Rabhi in #759
• Added local_files_only option on whisperx.load_model for offline mode by @RoqueGio in #867
• adding cache_dir to wav2vec2 by @bnitsan in #681
• feat: add basic installation test flow & restrict python versions by @Barabazs in #965
• chore: add build and release workflow by @Barabazs in #966
• fix: update README image source and enhance setup.py for long description by @Barabazs in #968
• docs: update installation instructions in README by @Barabazs in #969
• fix: add UTF-8 encoding when reading README.md by @xigh in #970
• chore: loosen ctranslate2 version restriction & bump whisperX version by @Barabazs in #971

New Contributors

• @moritzbrantner made their first contribution in #814
• @Hasan-Naseer made their first contribution in #875
• @ibombonato made their first contribution in #936
• @H4CK3Rabhi made their first contribution in #759
• @RoqueGio made their first contribution in #867
• @bnitsan made their first contribution in #681
• @xigh made their first contribution in #970

Full Changelog: v3.2.0...v3.3.0

v3.2.0

Device and Language Support

• added Korean wav2vec2 model by @Boulaouaney in #277
• Add Czech alignment model by @Thebys in #280
• Adding Norwegian Bokmål and Norwegian Nynorsk by @peregilk in #636
• Support language names in --language parameter. by @jkukul in #517
• Add align model for catalan language. by @davidmartinrius in #581
• add missing Cantonese in supported languages by @MahmoudAshraf97 in #617
• Add alignment model for Malayalam by @kurianbenoy in #585
• Added Romanian phoneme-based ASR model by @Majdoddin in #791
• added alignment for sk and sl languages by @jan-panoch in #852
• Add war2vec model for Vietnamese in #278
• Add Urdu model support for alignment by @abCods in #374
• chore(writer): Join words without spaces for ja, zh by @jim60105 in #440

Bug Fixes and Stability Improvements

• fix Unequal Stack Size VAD error by @m-bain in #281
• fix: Bug in type hinting by @VisionOra in #294
• pin faster whisper by @sorgfresser in #474
• Fix repeat transcription on different languages and proper suppress_numerals use by @Joemgu7 in #395
• fix writer fail on segments 0 by @sorgfresser in #429
• fix missing speaker prefix by @invisprints in #438
• fix: correct defaut_asr_options with new options (patch 0.8) by @remic33 in #458
• Fixes --model_dir path by @canoalberto in #648
• fix: force ctranslate to version 4.4.0 by @Barabazs in #946
• fix: update faster-whisper dependencies by @cococig in #716
• fix: ZeroDivisionError when --print_progress True by @mvoggu in #494
• Minor fixes for word options and subtitles by @amolinasalazar in #549
• fix unboundlocalerror by @sorgfresser in #554
• Fix: Allow vad options to be configurable by passing to FasterWhisperPipeline and merge_chunks. by @abettke in #507
• fix minimum input length for torch wav2vec2 models by @MahmoudAshraf97 in #510
• fix(diarize): key error on empty track by @characat0 in #518
• pip compliance for git+ installs by @spbisc97 in #603

Documentation Updates

• adds link to whisperX medium on replicate.com by @CaRniFeXeR in #431
• Document --compute_type command line option by @dotgrid in #430
• adding link to Replicate demo by @daanelson in #352
• fix: typo in error message by @zamoshchin in #493
• Fix link in README.md by @jimregan in #668
• Update README.md by @valentt in #509
• Add a special note about Speaker-Diarization-3.0 in readme by @kaihe-stori in #521
• Update README to correct speaker diarization version link by @gillens in #618
• Update README.md by @mlopsengr in #630
• fix link by @M0HID in #605
• Remove torchvision from README by @baer in #378

Miscellaneous Changes

• move model to assets by @m-bain in #945
• Update alignment.py by @Ayushi-Desynova in #418
• Update alignment.py by @awerks in #427
• push contributions from main by @m-bain in #290
• make diarization faster by @davidas1 in #400
• Add device_index option by @sorgfresser in #266
• Add transcribe keywords by @sorgfresser in #269
• Added download path parameter. by @prameshbajra in #284
• Suppress numerals by @m-bain in #303
• Add Audacity export by @Ca-ressemble-a-du-fake in #309
• Update transcribe.py -> small change in batch_size description by @mabergerx in #382
• Suggest using pytorch-cuda 11.8 instead of 11.7 by @tijszwinkels in #255
• feat: Add merge chunks chunk_size as arguments. by @jim60105 in #445
• A solution to long subtitles and words without timestamps by @awerks in #459
• chore(writer): improve text display(ja etc) in json file by @darwintree in #472
• add faster whisper threading by @sorgfresser in #473
• Pyannote3 by @remic33 in #492
• Update alignment.py by @piuy11 in #487
• Pass patience and beam_size to faster-whisper. by @jkukul in #527
• remove the minimum length for alignment and print the failing segment by @MahmoudAshraf97 in #529
• Update setup.py to use pyannote.audio version with working GPU by @wuurrd in #531
• Update setup.py to download pyannote depending on platform by @justinwlin in #541
• Drop ffmpeg-python dependency and call ffmpeg directly. by @hidenori-endo in #570
• no align based on space by @sorgfresser in #556
• Update asr.py and make the model parameter be used by @kaka1909 in #580
• Move load_model after WhisperModel by @DougTrajano in #584
• Update pyannote to 3.1.0 by @remic33 in #586
• support for large-v3 by @MahmoudAshraf97 in #599
• Added option to load Custom VAD model to load model method by @Swami-Abhinav in #654
• Update pyannote to v3.1.1 to fix a diarization problem (and diarize.py) by @santialferez in #646
• Get rid of numeral_symbol_tokens variable in printed message by @KossaiSbai in #669
• Add Replicate large-v3 demo by @victor-upmeet in #703
• local vad model by @m-bain in #944
• Feat: add new align models - SHORT by @Equipo45 in #922
• Update alignment.py by @peregilk in #687

Full Changelog: v3.1.1...v3.2.0

3.1.1

• translate functionality added
• fix word timestamp bug (words no longer have consecutive timestamps)