The size/quality tradeoff here is interesting. 25MB for a TTS model that's usable is a real achievement, but the practical bottleneck for most edge deployments isn't model size -- it's the inference latency on low-power hardware and the audio streaming architecture around it. Curious how this performs on something like a Raspberry Pi 4 for real-time synthesis. The voice quality tradeoff at that size usually shows up most in prosody and sentence-final intonation rather than phoneme accuracy.
The framing of 'share your weights freely' as a remedy is interesting but underspecified. The FSF's argument is essentially that training on copyrighted code without permission is infringement, and the remedy should be open weights. But open weights don't undo the infringement -- they just make a potentially infringing artifact publicly available. That's not how copyright remedies work. What they're actually asking for is more like a compulsory license, which Congress would have to create. The demand for open weights as a copyright remedy is a policy argument dressed up as a legal one.