BACK_TO_FEEDAICRIER_2
Fish Audio S2 Pro underwhelms, MOSS-TTS next
OPEN_SOURCE ↗
REDDIT · REDDIT// 19d agoNEWS

Fish Audio S2 Pro underwhelms, MOSS-TTS next

A Reddit thread says Fish Audio S2 Pro still sounds robotic even with emotion tags and notes that the model is licensed for research and non-commercial use unless you arrange a separate commercial license.

// ANALYSIS

Fish Audio feels like a strong demo trapped by a weak shipping story: if the voice still sounds flat and the license limits commercial use, the benchmark win doesn't matter much.

  • Fish Audio's own license allows research and non-commercial use, with commercial use requiring a separate license.
  • MOSS-TTS is the cleaner open-source bet: Apache-2.0, 20 languages, zero-shot cloning, long-form stability, and code-switching.
  • Qwen3-TTS is worth testing if you care about voice design and latency; the official repo advertises voice design, rapid cloning, and 97 ms streaming.
  • If emotion tags still collapse into robot voice, the real problem is usually conditioning, data, or decoding controls, not the whole TTS category.
// TAGS
speechaudio-genopen-sourcefish-audio-s2-promoss-ttsqwen3-tts

DISCOVERED

19d ago

2026-03-24

PUBLISHED

19d ago

2026-03-23

RELEVANCE

8/ 10

AUTHOR

FluffyMacho