How to save txt vtt and srt outputs? how to set beam_size, initial_prompt, best_of and other parameters

[FurkanGozukara]

I have checked main page and kaggle and there is no example of these

In reguler I was doing like below

For whisper jax how can I do?

        result = model.transcribe("../input/whisper2/lecture_"+str(lectureId)+".mp3",language="en",beam_size=10,initial_prompt="Welcome to the Software Engineering Courses channel.",best_of=10,verbose=True,temperature=0.0)

        # save SRT

        language = result["language"]
        sub_name = f"/kaggle/working/lecture_"+str(lectureId)+".srt"
        with open(sub_name, "w", encoding="utf-8") as srt:
            write_srt(result["segments"], file=srt)

        # Save output
        writing_lut = {
            '.txt': whisper.utils.write_txt,
            '.vtt': whisper.utils.write_vtt,
            '.srt': whisper.utils.write_txt,
        }

[sanchit-gandhi]

Hey @FurkanGozukara,

To save the transcriptions as a .txt file, you can do the following:

pred_str = pipeline(...)

with open("output.txt", "w") as text_file:
    text_file.write(pred_str)

This will save your predictions to a file called output.txt.

We could definitely extend this to saving .vtt/.srt files. To do so, we'd need to port whisper/utils.py to Whisper JAX.

[ruimaia]

Hi, @sanchit-gandhi. Thanks for the amazing work you've done.

Regarding the decoding options such as beam_size, initial_prompt, etc. it does seem that it is not possible to change them. Do you confirm? Are there any plans to allow to tune those parameters in the future?

[rRobis]

Hey @FurkanGozukara,

To save the transcriptions as a .txt file, you can do the following:

pred_str = pipeline(...)

with open("output.txt", "w") as text_file:
    text_file.write(pred_str)

This will save your predictions to a file called output.txt.

We could definitely extend this to saving .vtt/.srt files. To do so, we'd need to port whisper/utils.py to Whisper JAX.

Please implement srt export, i would be very thankful.

[tropanets]

Guys, please provide the code to export in the SRT format. I will be grateful.

[nhan000]

is there any update on the SRT export please?

[FurkanGozukara]

ye i didnt spend any time because it lacks this most important feature

srt and vtt output like original whisper which i use daily

[troublesprouter]

Any updates on enabling temperature setting?

[CyberAndrew]

Hey @FurkanGozukara,

To save the transcriptions as a .txt file, you can do the following:

pred_str = pipeline(...)

with open("output.txt", "w") as text_file:
    text_file.write(pred_str)

This will save your predictions to a file called output.txt.

We could definitely extend this to saving .vtt/.srt files. To do so, we'd need to port whisper/utils.py to Whisper JAX.

Hi. When you will port "utils.py" to gift us ability to generate .srt files? I am very need it, but I'm not good at programming.

[sovdevs]

agree. Keen on being able to drop back into whisper to write output to SRT, VTT etc.

[rRobis]

Hey @FurkanGozukara,

To save the transcriptions as a .txt file, you can do the following:

pred_str = pipeline(...)

with open("output.txt", "w") as text_file:
    text_file.write(pred_str)

This will save your predictions to a file called output.txt.

We could definitely extend this to saving .vtt/.srt files. To do so, we'd need to port whisper/utils.py to Whisper JAX.

Hi!

I just saw this kaggle notebook (i didn't test it), maybe you can take a look and take the srt implementation from it?

https://www.kaggle.com/code/idealim/mp3-to-text-and-subtitles-with-whisper-jax/notebook

[rRobis]

@sanchit-gandhi

[Philipp-Sc]

@rRobis thanks a lot for sharing the code, the save_as_srt function works like a charm.