Is it possible to include punctuation in the transcription?
Hello,
First of all, thank you very much for providing this model — the accuracy is truly impressive!
As a Chinese user, I'm used to using Whisper for audio transcription, where I could add punctuation by setting the initial_prompt parameter. I tried using the same approach with this model via the CLI, but it seems punctuation is not included in the output, even when I specify initial_prompt.
Here is my command:
whisper C:\...\..\test.wav ^
--model breeze-asr-25 ^
--language zh ^
--device cuda ^
--output_format srt ^
--initial_prompt "嗯...呃，你好?"
Punctuation would significantly improve the readability and usability of the transcription results.
Is there a recommended way to enable punctuation support with this model?
Thank you again for your excellent work, and I really appreciate any guidance you can offer.
Hi,
Thank you for trying out Breeze ASR 25. Punctuation is unfortunately not supported in this model. Punctuation caused instability in our initial testing, so we replaced all punctuations with spaces.
If this is a necessity, you can try to finetune it with a little data, or use BERT-based methods to do post ASR punctuation insertions.
Understood. Really appreciate your answer!