1 Hour Audio Transcription in 1 Minute using only 1GB VRAM - whisper.cpp
ฝัง
- เผยแพร่เมื่อ 21 ต.ค. 2024
- In this video I download a 2 hour long audio from a youtube a creative commons video. I use quantized q5_0 whisper large-v3-turbo model here. It only takes 1 GB VRAM. Also show how to transcribe an hour long video in a minute and generate an automated video with text from transcript at exact timestamps. I enable Attention with "-fa" otherwise it takes a few hundred more MB memory.
I forgot compare the non-quantized full version whisper large-v3-turbo memory usage and speed in this video. It will be available in next video.
In my previous video I have shown fast whisper.cpp build process. Model and code links below.
Links:
www.patreon.co...
/ compactai
huggingface.co...
github.com/gge...