Outputting SRT format from Python

[ehobson]

The example provided on the repository page shows usage of the print result function:

import whisper

model = whisper.load_model("base")
result = model.transcribe("audio.mp3")
print(result["text"])

This prints an unordered block of text in the python console window. How can a timestamped SRT or TXT file be produced using the python console, the same as it would be produced from the main OS environment?

[glangford]

See utils.py. I haven't called it directly, but for SRT files the function you are looking for are write_srt(). See the example usage in the docstring here

whisper/whisper/utils.py

Line 64 in ec1b34b

def write_srt(transcript: Iterator[dict], file: TextIO):

[VimWei]

output_format="all"
output_dir="./"
output_writer = whisper.utils.get_writer(output_format, output_dir)
output_writer(result, audio_filename)

[glangford]

@VimWei Sorry, I misunderstood your question.

The latest version of utils.py has been refactored and there is an extra argument passed to the writer. The shortest way to write a .srt would be something like this:

from whisper.utils import get_writer

result = model.transcribe( ... )
writer = get_writer("srt", ".") # get srt writer for the current directory
writer(result, audio_file_name, {}) # add empty dictionary for 'options'

The subtitles writer could be given formatting options (rather than the empty directory), see the code here:

whisper/whisper/utils.py

Line 104 in 248b6cb

raw_max_line_width: Optional[int] = options["max_line_width"]

Let me know if that works for you.

[VimWei]

Thanks so much! It works.

[JD144-git]

@glangford
Hi!, maybe it happened some issues with 'srt' part. I'm working on it, and python don't let me advance. Can you help me?

[glangford]

@JD144-git
Try something like this

srt_writer(result, audio, {"max_line_width":50, "max_line_count":1, "highlight_words":False} )

[kevinsaracho]

help