How to make whisper use punctuations more correct?

[zeushera140]

For some audios, it uses Punctuations quite good, but for some, it sometimes does not use Punctuations for a long long sentence, especially for some cases it should use period to end this sentence, it did not.

I asked GPT-4 how to make it use Punctuations more correct, there are 2 ways:

1. Use initial prompt, I tried, but failed. It may work for the beginning of audio, but not working well for the whole file.
2. Use gpt to fix the generated result, I have not used this way, but gpt-4 is too expensive for doing this, and gpt-3.5 may be good. However, this way is also a little complex for this job, cause I have to write specific prompt for specific audio.

I also tried whisperx, it tried to end the sentence too often, while usually cut the sentence in the middle.

Any other solutions？

Thanks.

[glangford]

See these past discussions, and prefer the use of large-v2 rather than large-v3.

• No punctuation for the first 75 minutes of the video. What could be the error / bug? #194
• No punctuation for the first 75 minutes of the video. What could be the error / bug? #194 (comment)

[Purfview]

Any other solutions？

Standalone Faster-Whisper.

[zeushera140]

faster-whisper? I tried, not better.

[Purfview]

faster-whisper?

No. Standalone Faster-Whisper!

[zeushera140]

I thought Standalone Faster-Whisper use faster-whisper as backend model. I will try it.

[sjtu-hxj]

#194 (comment)

[zeushera140]

thanks, but I tried that, not always work.