A new edge variant of the SteerTTS text-to-speech model has been released, significantly reducing its size to 11.84 MB while maintaining word error rate. This smaller version achieves its size reduction through quantization-aware training of its acoustic model and vocoder to int4, and by using a pure NumPy text front end. Despite a minor drop in UTMOS score, the model retains most of its steerability features, allowing for control over parameters like rate, loudness, and pitch, and supports inline markup for stylistic changes. AI
IMPACT Enables more efficient deployment of advanced text-to-speech capabilities on resource-constrained devices.
RANK_REASON Release of a smaller, optimized version of an existing TTS model for edge devices.
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
