A genuine long-form breakthrough
Reviewers widely call 90-min, 4-speaker dialogue a real step change.
Reviews & reception
What benchmarks and the community say — and how it stacks up against others. A non-official digest of public opinion; tap a card for detail and source.
11 result(s)
Open, free, self-hostable
Praised for MIT licensing and avoiding subscription/privacy concerns.
Speaker consistency & naturalness
HN community credits its voice consistency and conversational naturalness.
67% rate expressiveness higher
An AllAboutAI survey: 67% of technical users rate it above Chatterbox-TTS.
Intonation still slips
An HN user: intonation is off on nearly every phrase, with robotic modulation.
Impressive, but not by today's bar
Some find it impressive vs years ago, yet uninspiring by current standards.
Language & feature limits
Reviews agree: EN/ZH-first, no overlapping speech, music or sound effects.
The 2026 models arrived quietly
Streaming ASR shipped with no launch post and, so far, no independent benchmarks.
MOS 4.5 — tops the benchmarks
Per MS's report, the 7B model's MOS beat ElevenLabs v3 and Gemini TTS.
vs ElevenLabs
Wins on long-form and cost; ElevenLabs still leads on polish and languages.
vs F5-TTS / XTTS
Better long-dialogue consistency; F5 still praised for single-utterance quality.