I think a lot of people have heard of OpenAI’s local-friendly Whisper model, but I don’t see enough self-hosters talking about WhisperX, so I’ll hop on the soapbox:
Whisper is extremely good when you have lots of audio with one person talking, but fails hard in a conversational setting with people talking over each other. It’s also hard to sync up transcripts with the original audio.
Enter WhisperX: WhisperX is an improved whisper implementation that automatically tags who is talking, and tags each line of speech with a timestamp.
I’ve found it great for DMing TTRPGs — simply record your session with a conference mic, run a transcript with WhisperX, and pass the output to a long-context LLM for easy session summaries. It’s a great way to avoid slowing down the game by taking notes on minor events and NPCs.
I’ve also used it in a hacky script pipeline to bulk download podcast episodes with yt-dlp, create searchable transcripts, and scrub ads by having an LLM sniff out timestamps to cut with ffmpeg.
Privacy-friendly, modest hardware requirements, and good at what it does. WhisperX, apply directly to the forehead.
Nice. I learned about different applications of whisper because I’m a degenerate.
I’ve also used it in a hacky script pipeline to bulk download podcast episodes with yt-dlp, create searchable transcripts, and scrub ads by having an LLM sniff out timestamps to cut with ffmpeg.
This is genius. Could you appify this and I’ll pay you in real or pretend currency as you prefer
I’ve found it great for DMing TTRPGs — simply record your session with a conference mic, run a transcript with WhisperX, and pass the output to a long-context LLM for easy session summaries. It’s a great way to avoid slowing down the game by taking notes on minor events and NPCs.
Okay that’s just crazy. ;)
Probably not that hard to build a simple flask frontend around it.
Automatically processing files in an S3/WebDAV directory would also be useful.
What would be some use cases for WhisperX? I’m struggling to envision how I would use that in a selfhosting/homelabbing environment.
Long videos or voice notes where you’re usually just looking for a small snippet.
half sarcastic but the overall premise of rigging something in to a local voice assistant, when an arguement starts “Ok nabu record this conversation”. then 2 weeks later on another arguement… “OK nabu search our last arguement for the cabinet”. Would be like having a court transcriber on call.
I have a lady friend that does quite a good enough job of that. LOL
‘You remember back in 1979…it was a Friday at 2:11 PM, and you said…’ ‘Babe, I don’t remember what I had for breakfast yesterday.’
Does she do it for her fuckups, though?
What kind of stupid-ass question is that? LOL All kidding aside, she’s a good soul. We’re not married, we’ve just know each other for 45+ years. It just kind of clicked. She lives in her house, and I in mine, and we get together as often as possible.
Hmm… Would be interesting to find out what kind of effect that has on the average marriage or relationship 😅
I mean, I’d imagine probably not a good one :) Somehow I imagine asking the AI to record a conversation, is an instant arguement escalator… as is asking to read the facts back, and usually the topic would be switched rather than one side admitting their fault in the conversation.
Actually I think there’s a black mirror episode on roughly that (not a device for recording audio when asked, but everyone having a chip in their head that automatically records their memories, and a huge fight when a husband discovers his wife deleted a few hours of recordings.
“You love the robot more than me!” 💔️
Likely everyday stuff… Meeting minutes, phone or video conferences and such…
I guess that’s why I am having difficulty coming up with a use case. I mean, I walk around the lab talking to myself all day long, but I think it’d be a bad idea to have a record of all those conversations. lol
If you don’t have to sit through a bunch of ‘meetings that could have been emails’ on a daily basis, you likely won’t have a use case for it.
But in my last job I was a systems engineer for a web development company. I had to be included on all of the dev calls in case an infrastructure question came up that I needed to answer, and so I was vaguely aware of what the devs were doing.
This software would have been a lifesaver, because my ADHD doesn’t let me listen to stuff like that for a straight hour or two.
That is cool! I’ve been wanting I’ve wanted to use a model like this but haven’t really looked.
Are you self hosting the long context llm, of do what are you using?
Context lengths are what kill a lot of my local llm experiments.
Are you self hosting the long context llm, of do what are you using?
I did a lot of my exploration back when GPT4 128K over API was the only long-context game in town.
I imagine the options are much better these days between Llama 3/4, Deepseek, and Qwen — but haven’t tried them locally myself.