1 Hour Audio Transcribed in 1 Minute with 1GB VRAM - whisper.cpp

In this video I transcribe an 1 hour long video with whisper.cpp using whisper v3 large turbo q5 quantized model requiring only 1GB VRAM memory.

In my previous video I have shown fast whisper.cpp build process. Model and code links below. In this video I use an AI generated voice. I have used voice from VCTK VITS model for voicing this video

Topics Covered:
- Using yt-dlp tool to download 2 hour long youtube audio from a video in jupyter notebook using python.
- Select video, audio streams, quality and formats to download with yt-dlp video downloader.
- Running server for whisper.cpp and uantized q5_0 whisper large-v3-turbo model, then getting VTT format speech transcription.
- Using FFMPEG to convert video to audio from a given to another timestamp.
- Additionally, I show using flash attention "-fa" flag to speed up inference.
- Finally, I show to generate an automated video from Speech to Text generated by whisper model using python.

Get Involved:
If you found this helpful, consider liking, subscribing, and sharing with others as I delve into more AI-related topics. Share your thoughts on the video in the comments! Let’s explore the global potential of AI together as a part of the worldwide technology and AI community.

For those who want to support my work, consider visiting my Patreon for additional resources and codes: https://www.patreon.com/CompactAI

Links:
https://www.patreon.com/CompactAI
https://www.patreon.com/posts/1-hour-audio-in-114479515
https://huggingface.co/ggerganov/whisper.cpp
https://github.com/ggerganov/whisper.cpp

Hashtags: #whisper #speechtotext #videogeneration #jupyternotebook #python #ffmpeg #videodownloader #openai #quantization #vscode #tutorial #programming #windows

Language: English.

Видео 1 Hour Audio Transcribed in 1 Minute with 1GB VRAM - whisper.cpp канала Compact AI