ccstan99
commented
10 months ago Maybe we can use LangChain document loaders in conjuction with their whisper parser? https://python.langchain.com/docs/integrations/document_loaders/youtube_audio
# ! pip install yt_dlp
# ! pip install pydub
from langchain.document_loaders.generic import GenericLoader
from langchain.document_loaders.parsers import OpenAIWhisperParser
from langchain.document_loaders.blob_loaders.youtube_audio import YoutubeAudioLoader
url="https://www.youtube.com/watch?v=jGwO_UgTS7I"
save_dir="docs/youtube/"
loader = GenericLoader(
YoutubeAudioLoader([url],save_dir),
OpenAIWhisperParser()
)
docs = loader.load()
We are currently using transcripts directly from YouTube. Many are low quality, without proper capitalization & punctuation and odd formatting. There are Otter transcripts at https://drive.google.com/drive/u/0/folders/12FFjcP_60rq3qgq_Wl7HDWIjEA7p_3T_ for many of the YouTube videos from the previous ARD but we'll need to reconcile the transcripts with the associated URL.
For new videos, we may want to consider integrating Whisper transcripts into the workflow, i.e. https://huggingface.co/spaces/SteveDigital/free-fast-youtube-url-video-to-text-using-openai-whisper