Confirmed
Whissle AI releases SteerTTS edge, a steerable TTS model at 11.84 MB
Int4 quantized acoustic model and vocoder run at about 19x realtime on a CPU with 4 threads. English only with 200 voices and no voice cloning on device. Seed-TTS WER stays at 0.96% while UTMOS falls from 4.20 to 4.11, per the developer's local run.
Every outlet that ran it 1 article
r/MachineLearningCommunityLead
We shrank our steerable small TTS to 11.84 MB. WER is unchanged, UTMOS drops ~0.1 [p]