heimoshuiyu/faster-whisper
1
0
Fork 0
Code Issues Pull Requests 1 Packages Projects Releases Wiki Activity

Add support for distil-large-v3 (#755)

Browse Source 
* add distil-large-v3

* Update README.md

* use fp16 weights from Systran
This commit is contained in:
Sanchit Gandhi
2024-03-26 13:58:39 +00:00
committed by GitHub
parent 1eb9a8004c
commit a67e0e47ae
2 changed files with 16 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

23
README.md
View File

@@ -159,18 +159,25 @@ for segment in segments:
segments, _ = model.transcribe("audio.mp3") segments, _ = model.transcribe("audio.mp3")
segments = list(segments) # The transcription will actually run here. segments = list(segments) # The transcription will actually run here.
``` ```
### Faster-distil-whisper ### Faster Distil-Whisper
For usage of `faster-distil-whisper`, please refer to: https://github.com/guillaumekln/faster-whisper/issues/533
The Distil-Whisper checkpoints are compatible with the Faster-Whisper package. In particular, the latest [distil-large-v3](https://huggingface.co/distil-whisper/distil-large-v3)
checkpoint is intrinsically designed to work with the Faster-Whisper transcription algorithm. The following code snippet
demonstrates how to run inference with distil-large-v3 on a specified audio file:
```python ```python
model_size = "distil-large-v2" from faster_whisper import WhisperModel
# model_size = "distil-medium.en"
model = WhisperModel(model_size, device="cuda", compute_type="float16")
segments, info = model.transcribe("audio.mp3", beam_size=5,
language="en", max_new_tokens=128, condition_on_previous_text=False)
model_size = "distil-large-v3"
model = WhisperModel(model_size, device="cuda", compute_type="float16")
segments, info = model.transcribe("audio.mp3", beam_size=5, language="en", condition_on_previous_text=False)
for segment in segments:
print("[%.2fs -> %.2fs] %s" % (segment.start, segment.end, segment.text))
``` ```
NOTE: Empirically, `condition_on_previous_text=True` will degrade the performance of `faster-distil-whisper` for long audio. Degradation on the first chunk was observed with `initial_prompt` too.
For more information about the distil-large-v3 model, refer to the original [model card](https://huggingface.co/distil-whisper/distil-large-v3).
### Word-level timestamps ### Word-level timestamps

1
faster_whisper/utils.py
View File

@@ -25,6 +25,7 @@ _MODELS = {
"distil-large-v2": "Systran/faster-distil-whisper-large-v2", "distil-large-v2": "Systran/faster-distil-whisper-large-v2",
"distil-medium.en": "Systran/faster-distil-whisper-medium.en", "distil-medium.en": "Systran/faster-distil-whisper-medium.en",
"distil-small.en": "Systran/faster-distil-whisper-small.en", "distil-small.en": "Systran/faster-distil-whisper-small.en",
"distil-large-v3": "Systran/faster-distil-whisper-large-v3",
} }