Question for folks who work a lot with STT models - What is your favorite model that supports word-level timestamps, has good dysfluency detection (whisper isn't great), and is also supported by transformers.js?
That's a great question - these are sold as NFTs, and I think you only get the art, not the input parameters used to generate the art. I really like that idea though
Speech MCP Server - uses Kokoro TTS under the hood to give your LLM the ability to speak. I use it to have Cursor agents notify me when long running tasks are complete.
I've been working on a free, in-browser "pre" video editor. Upload your clip, use a transcript interface to cut it down to the takes and salient bits you want to keep, then export your cuts to FCP or Resolve to complete your editing. This tool saves me about 25-30% of my editing time.
Uses transformers.js & WebGPU for running transcription, so it's pretty fast. It's still a bit rough around the edges, so I'm looking for feedback.
I think this might be really cool, but when I tried clicking "playback actions" (unclear that this button means "play" and won't just take me to a list of actions) there was no sound.