QUT-Digital-Observatory / youscribe

Youtetube downloader/transcriber for Youte+

Transcribe YouTube videos using Whisper models

Adopts faster_whisperer, a cTransformer's based model for faster transcription.

Usage

from youscribe import transcribe

transcript = transcribe(url="https://www.youtube.com/watch?v=9bZkp7q19f0")

transcript.text()

Prompting

By default, the video title and description are used as prompts to the transcription model. But you can also specify your own prompt:

transcript = transcribe(
    url="https://www.youtube.com/watch?v=9bZkp7q19f0",
    prompt="Enter prompt here"
)

You can also choose not to include prompt by setting prompt=False.

transcript = transcribe(
    url="https://www.youtube.com/watch?v=9bZkp7q19f0",
    prompt=False
)

Working with WhisperTranscript objects

The transcribe() function, if executed successfully, will return a WhisperTranscript object. You can view the transcript as plain text, SRT-formatted text, or a Python dictionary.

transcript = transcribe(
    url="https://www.youtube.com/watch?v=9bZkp7q19f0",
    prompt=False
)

transcript.text()
transcript.srt()
transcript.json()
transcript.segment

Customise Whisper model

In the transcribe function, you can pass your own custom Whisper model:

from youtescribe import WhisperTranscriber
from youtescribe import models

custom_transcriber = WhisperTranscriber(model_size = models.TINY_EN, cpu_threads=6, device="auto")

transcript = transcribe(
    url="https://www.youtube.com/watch?v=9bZkp7q19f0",
    transcriber=custom_transcriber
)
transcript.text()