Skip to content

Recipe: transcripts → LLM / RAG chunks

Split a transcript into ~N-second chunks that keep a timestamp link back to the video.

from ytscrape import YouTube

def chunk(transcript, seconds: float = 60.0):
    buf, start = [], None
    for s in transcript.snippets:
        start = s.start if start is None else start
        buf.append(s.text)
        if s.start + s.duration - start >= seconds:
            yield {"start": start, "text": " ".join(buf)}
            buf, start = [], None
    if buf:
        yield {"start": start, "text": " ".join(buf)}

with YouTube() as yt:
    tr = yt.transcript("dQw4w9WgXcQ", languages=["en"])

docs = [
    {**c, "url": f"https://youtu.be/{tr.video_id}?t={int(c['start'])}"}
    for c in chunk(tr)
]
# embed docs[i]["text"] with your favourite embedding model / vector store

import asyncio
from ytscrape import AsyncYouTube

async def main():
    async with AsyncYouTube() as yt:
        return await yt.transcript("dQw4w9WgXcQ", languages=["en"])

tr = asyncio.run(main())

Handle TranscriptsDisabled and NoTranscriptFound for videos without captions.