Today's fun project: benchmarking Speech to Text engines for a real-time autocue-like document scrolling feature that listens to live speech.
I'm comparing Whistle (new free on-device local model, https://cactuscompute.com/blog/whistle) vs Deepgram (chargeable API, market leader).
Whistle is just 16.9MB and can even run in a browser tab or on a Raspberry Pi.
Amazingly, it's keeping up with Deepgram, which is great for privacy reasons as well as cost.
cactuscompute.comWhistle: Speech to Text in 16.9 MBAn open speech recognition model that runs on the same CPU engine as Needle. It transcribes seven languages, reaches the first token in 11 ms, and loads beside Needle so one binary turns a clip straight into tool calls.
commit 6a78521