Clean implementation. One thing I always look for: how does this degrade when things go wrong? Good error handling is what separates weekend projects from tools people actually use.
(responding to dead sibling) As someone who frequently replies to people's questions with "Great question!" or "That's a good point!," I'm a little concerned people are going to start thinking I'm an LLM.
Sorry if this is a dumb question, but could this be modified to create tab from a live performance? I improvised a solo with my band recently and it was a big hit and I'm feeling too lazy to transcribe it.
Cool project! I built something similar a while back and then kept adding to it: https://github.com/rmusser01/tldw_server
It’s now grown quite a bit past the original goal of transcribing and summarizing conference talks for me.
Can you explain? yt-dlp works fine, especially if you just use your browser cookies. the fact you can select auto-subtitles, language-specific subtitles and the fact it also works in other websites like patreon is what made me realize how well-built yt-dlp is.
Could be related to ips - i think it works well at my house - but when I am working from a specific cafe, it does not. But yeah seems to be getting more erratic lately.
Yes. Most videos on Youtube have captions. OpenBrief is basically a wrapper of yt-dlp so it can support other video streaming services and transcribe them if it doesn't have captions. One of my use cases is importing audio recordings. Local transcription model can do many things actually. The title might be over simplified.
Clean implementation. One thing I always look for: how does this degrade when things go wrong? Good error handling is what separates weekend projects from tools people actually use.
(responding to dead sibling) As someone who frequently replies to people's questions with "Great question!" or "That's a good point!," I'm a little concerned people are going to start thinking I'm an LLM.
Sorry if this is a dumb question, but could this be modified to create tab from a live performance? I improvised a solo with my band recently and it was a big hit and I'm feeling too lazy to transcribe it.
That's cool ... will definitely try that. I know there are many tools like that but when it gives good results in the end I'll use it.
Cool project! I built something similar a while back and then kept adding to it: https://github.com/rmusser01/tldw_server It’s now grown quite a bit past the original goal of transcribing and summarizing conference talks for me.
This feels like the 'Obsidian for video'. I your approach: local-first, ownership of files, and composable AI instead of another cloud subscription.
Curious how you handle long-context transcripts locally. Are you chunking + retrieval, or relying on large-context hosted models through BYO APIs?
Nice work. It's refreshing to see a simple and local-first Tauri app instead of another one CLI/TUI tool.
isn't this quite similar to https://github.com/kouhxp/yapsnap just with a llm for summarisation?
Within the past year, yt-dlp has ceased to work reliably, it seems Google is cracking down.
Can you explain? yt-dlp works fine, especially if you just use your browser cookies. the fact you can select auto-subtitles, language-specific subtitles and the fact it also works in other websites like patreon is what made me realize how well-built yt-dlp is.
Could be related to ips - i think it works well at my house - but when I am working from a specific cafe, it does not. But yeah seems to be getting more erratic lately.
Looks interesting, starred, thanks.
local-first. Love it..
Don’t YouTube videos already have a transcript?
Yes. Most videos on Youtube have captions. OpenBrief is basically a wrapper of yt-dlp so it can support other video streaming services and transcribe them if it doesn't have captions. One of my use cases is importing audio recordings. Local transcription model can do many things actually. The title might be over simplified.
It's way worse than what you can get locally via whisper. Not sure why, perhaps Google legally can't use whisper, or can't spare the compute.