Phonon TTS, a high-quality on-device speech synthesis model, was introduced, and speech research the same week clustered around streaming and conversational processing.
Hugging Face featured Phonon TTS, described as running on-device and topping benchmarks. In the same week arXiv papers covered a streaming question-answering benchmark, policy learning for simultaneous speech translation, and TTS distillation for better pronunciation control.
An on-device option reduces dependence on the cloud, which can matter for latency and privacy. Research is also leaning toward real-time processing and pronunciation accuracy, so release and papers point the same way. Note: the benchmark claim comes from the announcement and is not independently verified here.
Watch for real-world quality and hardware requirements, licensing terms, and adoption reports from developers.