As reported in recent research announcements, the landscape of speech and voice models is rapidly evolving. According to published reports, Fish Audio has released Live 3, described as a full-duplex voice model capable of reacting in 186 ms (Evidence 1). Additionally, Zyphra introduced ZONOS3-Mini, an open-source text-to-speech (TTS) model featuring 100M active parameters (Evidence 3). Furthermore, Hugging Face blog updates highlight Phonon TTS, which launched high-quality on-device speech reported to top benchmarks (Evidence 18).
*Analysis:* Based strictly on these published summaries, these releases indicate a growing emphasis on low-latency interactive voice capabilities and efficient open-source on-device speech generation. No hands-on testing or independent verification has been performed by this publication.