Whisper YouTube Transcriber

Transcribes an audio source to text using the OpenAI's Whisper API. The result is saved in the timestamps/ folder as a Markdown file. The input can be URL to a video source, it will use yt-dlp to download the video and proceed with transcribing.

Result Folders

• audios/ folder contains extracted audio files from YouTube videos when the input is a YouTube URL. Large audio files are split into smaller chunks are also saved here.
• jsons/ folder contains responses from Whisper API transcribing/translating the audio files. The data follows the schema in transcription.schema.json. Some non-speech audio segments may be falsely transcribed as speech by the Whisper model. These segments are filtered out and saved into json/*-no_speech.json and json/*-speech.json files.
• timestamps/ folder contains the resulting Markdown files, that includes transcription and their timestamps. The timestamps link to the YouTube video at the matching times when the input when the input is a YouTube URL or the input file has the source YouTube URL embedded to the metadata under format:tags:comment. The *-no_speech.json and *-speech.json files are also timestamped and saved here as timestamps/*-no_speech.md and timestamps/*-speech.md.

Usage

python main.py [-h] [-p PROMPT] [-l LANGUAGE] [-t] input

positional arguments:
  input                 Path or URL to the audio source to transcribe, in one of these formats: mp3, mp4, mpeg, mpga,
                        m4a, wav, or webm.

options:
  -h, --help            show this help message and exit
  -p PROMPT, --prompt PROMPT
                        A list of correct word spellings for problematic words.
  -l LANGUAGE, --language LANGUAGE
                        The language of the input audio. Supplying the input language in ISO-639-1 format will improve
                        accuracy and latency.
  -t, --translate       Translate the audio file to English.

Prompting

See OpenAI's documantation.