28 Commits
Author SHA1 Message Date
Purfview 2a3a857737 Fix CI failure caused by ubuntu-latest migration (#1686)
* Fix CI failure caused by ubuntu-latest migration
2026-11-02 11:19:10 +00:00
Purfview 9e472ae9de Raise an error for languages unsupported by a model (#1564)
* Raise an error for languages unsupported by a model

Close: https://github.com/SYSTRAN/faster-whisper/issues/1228
2026-10-16 12:12:55 +01:00
Purfview 47e2558a88 Raise an error for illegal values in clip_timestamps arg (#1654)
* Raise an error for illegal values in collect_chunks {input from clip_timestamps arg and and get_speech_timestamps]
2026-10-16 12:00:49 +01:00
Purfview 2486efd9e9 Verify that config.json has the required keys (#1673)
Verify that config.json has the required keys.
2026-10-16 11:45:24 +01:00
Purfview a52b026ac8 Tweak collect_chunks() (#1653)
A new group opens with missing "segments".

Not a bug in the API as "segments" metadata is not in use down stream.

Detected by some AI bot there: https://github.com/SYSTRAN/faster-whisper/pull/1491
2026-10-10 04:39:41 +01:00
Purfview 7b99be5376 Comment edit: Clarify chunk_length usage (#1536)
Added clarification on chunk_length usage after noticing some unexpected uses of this argument.

Some people even use it to shorten the text segments instead of post-processing the text. :)
2026-10-01 15:01:31 +01:00
PurfviewandMahmoud Ashraf ed9a06cd89 Adds new VAD parameters (#1386)
* Adds new VAD parameters

Adds new VAD parameters: 

min_silence_at_max_speech: Minimum silence duration in ms which is used to avoid abrupt cuts when max_speech_duration_s is reached.

use_max_poss_sil_at_max_speech: Whether to use the maximum possible silence at max_speech_duration_s or not. If not, the last silence is used.

* Style

* Update doc

* change min_speech_duration_ms (0 -> 250)

* Change min_speech_duration_ms to zero

Set minimum speech duration to zero for flexibility.

---------

Co-authored-by: Mahmoud Ashraf <hassouna97.ma@gmail.com>
2025-11-19 17:40:46 +03:00
Purfview 2eeafe05de Update Silero-VAD weights to v6.2 (#1390)
* Update Silero-VAD weights to v6.2

Overall slight quality improvement (no metrics update);
Higher stability on OOD / rare / strange / unique data;
Significant quality improvements on various known edge cases:

    Unusual voices
    Child voices
    Cartoon voices
    Muted voices
    Muted speech
    Lower quality phone calls

https://github.com/snakers4/silero-vad/releases/tag/v6.2

* Changes: tiny -> base in test_monotonic_timestamps()
2025-11-19 17:14:42 +03:00
Purfview cf42429f96 Remove "local_dir_use_symlinks" from download_model() (#1389)
* Remove "local_dir_use_symlinks" from download_model()

It's deprecated since huggingface_hub v0.23.0 and produce this warning:

>   /opt/hostedtoolcache/Python/3.9.24/x64/lib/python3.9/site-packages/huggingface_hub/utils/_validators.py:202: UserWarning: The `local_dir_use_symlinks` argument is deprecated and ignored in `snapshot_download`. Downloading to a local directory does not use symlinks anymore.

* Bump huggingface_hub requirement to v0.23
2025-11-18 21:59:01 +03:00
Purfview 9090997d25 Fix a typo (#1377) 2025-10-22 15:51:56 +03:00
Purfview 11fd8ab301 Fix neg_threshold (#1191) 2024-12-29 14:38:58 +03:00
PurfviewandMahmoud Ashraf 1b24f284c9 Reduce VAD memory usage (#1198)
Co-authored-by: Mahmoud Ashraf <hassouna97.ma@gmail.com>
2024-12-12 15:23:30 +03:00
Purfview f32c0e8af3 Make batched suppress_tokens behaviour same as in sequential (#1194) 2024-12-11 14:51:38 +03:00
Purfview 8327d8cc64 Brings back original VAD parameters naming (#1181) 2024-12-01 20:41:53 +03:00
Purfview b024972a56 Foolproof: Disable VAD if clip_timestamps is in use (#769)
* Foolproof: Disable VAD if clip_timestamps is in use

Prevent silly things to happen.
2024-04-02 18:20:34 +02:00
Purfview 8ae82c8372 Bugfix: code breaks if audio is empty (#768)
* Bugfix: code breaks if audio is empty

Regression since https://github.com/SYSTRAN/faster-whisper/pull/732 PR
2024-04-02 18:18:12 +02:00
Purfview 5090cc9d0d Fix window end heuristic for hallucination_silence_threshold (#706)
Removes the wishful heuristic causing more issues than it's fixing.

Same as https://github.com/openai/whisper/pull/2043

Example of the issue: https://github.com/openai/whisper/pull/1838#issuecomment-1960041500
2024-02-29 17:59:32 +01:00
Purfview 30d6043e90 Prevent infinite loop for out-of-bound timestamps in clip_timestamps (#697)
Same as https://github.com/openai/whisper/pull/2005
2024-02-22 09:48:35 +01:00
Purfview 52695567c9 Bumps up PyAV version to support Python 3.12.x (#679) 2024-02-20 17:31:07 +01:00
Purfview f144e4c83d Expands the note for distil-whisper (#659) 2024-01-28 21:48:40 +01:00
Purfview 3aec421849 Add: More clarity of what "max_new_tokens" does (#658)
* Add: More clarity of what "max_new_tokens" does
2024-01-28 21:40:33 +01:00
Purfview 00efce1696 Bugfix: Illogical "Avoid computing higher temperatures on no_speech" (#652) 2024-01-24 11:54:43 +01:00
Purfview 44f7e58947 Update whisper-standalone-win description in README.md (#508)
* Update whisper-standalone-win description in README.md
2023-12-14 13:03:46 +01:00
Purfview ebcfd6b964 Fix broken prompt_reset_on_temperature (#604)
* Fix broken prompt_reset_on_temperature

Fixing: https://github.com/SYSTRAN/faster-whisper/issues/603

Broken because `generate_with_fallback()` doesn't return final temperature.

Regression since PR356 -> https://github.com/SYSTRAN/faster-whisper/pull/356
2023-12-13 13:14:39 +01:00
Purfview 65094b779e Update info on cuBLAS and cuDNN libs in README.md (#513) 2023-11-27 12:12:47 +01:00
Purfview 1ce16652ee Adds DEBUG log message for prompt_reset_on_temperature (#399)
Produce DEBUG log message if prompt_reset_on_temperature threshold is met.
2023-08-04 09:06:17 +02:00
Purfview 857be6f621 Rename clear_previous_text_on_temperature argument (#398)
`prompt_reset_on_temperature` is more clear what it does.
2023-08-03 18:44:37 +02:00
Purfview 32b962bed8 Adds: whisper-standalone-win (#216) 2023-05-09 20:20:41 +02:00