Summary
According to a post on X by @FluidVoiceApp, the team released its speech-to-text model as open weights on Hugging Face. The model has 0.6B parameters and is about 600MB on device, and the post says it handles noisy cafés well and beats ElevenLabs and Deepgram on its own benchmarks. The same post points to kale-tts-1-nano from @NERDDISCO, a 7.5M-parameter, 14.8MB text-to-speech model with 10 voices, which the post says starts speaking in 90 ms on a plain CPU. These are the developers' own claims, and the snippet provided does not include independent testing.
AI-assisted summary based on the listed source.
What happened
@FluidVoiceApp released its speech-to-text model as open weights on Hugging Face: 0.6B parameters, about 600MB on device, which it says handles noisy cafés well and beats ElevenLabs and Deepgram on its own benchmarks. @NERDDISCO released kale-tts-1-nano, a 7.5M-parameter, 14.8MB text-to-speech model with 10 voices...
Why it matters
Developers building voice features could test small open models that run on device instead of depending on a hosted speech service. The noisy-environment and benchmark claims come from the developers themselves, so outside testing will matter before anyone relies on them. Watch for independent benchmarks and the full Hugging Face release details.
Signal Intelligence
Signal Strength 95%
Technical label WATCH
Public Interest 68
Category OPEN SOURCE
Reader Depth TECHNICAL
Signal Strength reflects source quality, relevance, freshness and evidence. Public Interest helps organize discovery; it is not proof of truth.
Public Interest components
Recognizable Entity Score 90
Practical Impact Score 46
Novelty Interest Score 94
Consequence Score 50
Curiosity Score 32
Shareability Score 73
Why this is here
VQV surfaced this signal because it is recent, relevant to AI Voice, connected to Scobleizer News Pulse (X).