From eaf38309a28216cad6823e1c8c0f52434d19d50a Mon Sep 17 00:00:00 2001 From: Bryant21 <5288648+bmalaski@users.noreply.github.com> Date: Sun, 4 Aug 2024 13:22:51 -0400 Subject: [PATCH 1/3] Update finetune.py --- finetune.py | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) diff --git a/finetune.py b/finetune.py index f4b11f8..6131991 100644 --- a/finetune.py +++ b/finetune.py @@ -1607,7 +1607,9 @@ if __name__ == "__main__": "large-v2", "large", "medium", - "small" + "small", + "distil-large-v3" + "distil-large-v2" ], scale=1, ) @@ -1722,7 +1724,7 @@ if __name__ == "__main__": 3. Provide at least 2 minutes of audio, preferably 5-10 minutes for better results.
4. Maximum Audio Length will ensure the audio samples do not exceed the set value in seconds. The AI model can only process so much audio per audio sample, per epoch.
5. Enter a unique Project Name if you wish to create a dedicated training folder. You will need to specify this name as you move through the interfaces and click the Refresh buttons where needed.
- 6. Select the Whisper model (large-v2 recommended for English use cases).
+ 6. Select the Whisper model (distil-large-v3 recommended for English use cases).
7. Choose your Dataset Language.
8. Adjust the Evaluation Split percentage (default 15% is suitable for most cases).
9. Optionally enable the BPE Tokenizer for custom tokenization, which can improve the model's ability to handle unique words, accents, or languages by creating a vocabulary tailored to your specific dataset. This is especially useful for non-standard speech patterns or languages with complex morphology.
From 94c9257deeab1d9abd7508f08b19651df8e31746 Mon Sep 17 00:00:00 2001 From: Bryant21 <5288648+bmalaski@users.noreply.github.com> Date: Sun, 4 Aug 2024 13:27:42 -0400 Subject: [PATCH 2/3] change default to distil --- finetune.py | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/finetune.py b/finetune.py index 6131991..309864c 100644 --- a/finetune.py +++ b/finetune.py @@ -1601,15 +1601,15 @@ if __name__ == "__main__": ) whisper_model = gr.Dropdown( label="Whisper Model", - value="large-v2", + value="distil-large-v3", choices=[ - "large-v3", - "large-v2", - "large", - "medium", - "small", - "distil-large-v3" - "distil-large-v2" + ("large-v3", "large-v3"), + ("large-v2", "large-v2"), + ("large", "large"), + ("medium", "large"), + ("small", "large"), + ("distil-large-v3 (en only)", "distil-large-v3"), + ("distil-large-v2 (en only)", "distil-large-v3"), ], scale=1, ) From 6b43c40a8c81551f488e9ff33c6147664f80ec8f Mon Sep 17 00:00:00 2001 From: Bryant21 <5288648+bmalaski@users.noreply.github.com> Date: Sun, 4 Aug 2024 13:33:53 -0400 Subject: [PATCH 3/3] Update finetune.py --- finetune.py | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/finetune.py b/finetune.py index 309864c..c7d50cf 100644 --- a/finetune.py +++ b/finetune.py @@ -1609,7 +1609,7 @@ if __name__ == "__main__": ("medium", "large"), ("small", "large"), ("distil-large-v3 (en only)", "distil-large-v3"), - ("distil-large-v2 (en only)", "distil-large-v3"), + ("distil-large-v2 (en only)", "distil-large-v2"), ], scale=1, )