Fish Audio's open-weights Drama 3 tops TTS arenas with best-in-class pronunciation robustness
ArtificialAnlys · x · 2026-10-29
Fish Audio released Drama 3, and Artificial Analysis evaluated it across both Speech Arenas and its Pronunciation Robustness benchmark:
- Provider Voice: Elo 1278 across 1,512 appearances — #6 overall, #1 among open weights models, 38 points above its own Drama 3 Preview (1240) and 50 ahead of Breeze TTS 2 (1228)
- Controlled Voice: 1085, #1 open weights, ahead of Ant Group's Ming-3.0-tts (1068)
- Pronunciation Robustness: 94.6% — #1 of all models, ahead of Ming-3.0-tts and Drama 3 Preview (both 93.2%)
- Speed: 66.6 chars/sec, behind Eleven v4 (70.0) but ahead of Drama 3 Preview (60.7)
- Pricing: $15/1M characters, well below every model ranked above it (Eleven v4 costs $80/1M)
Released as open weights under the Fish Audio Community License.
Related event: Fish Audio open-sources Drama 3, tops open-weight TTS rankings(5 posts)→
More from Multimodal
- "Darkness": a stunning dusk city flyover AI video demo — sdrowegnarts · 2026-10-30
- Falcon: Streaming Audio-Visual Speaker Extraction Cuts Params 91.9% and Latency 11× — _akhaliq · 2026-10-30
- Open-source StableGen 0.4 turns prompts into rigged, textured characters inside Blender — sakalond · 2026-10-30
- Grok appears to have access to Suno's MCP connector, gated by early access — koltregaskes · 2026-10-30
- A visual guide to multimodal generation: 50+ figures, free bonus chapter — MaartenGr · 2026-10-30
- MOSS-Image vs SenseNova U1.5 Lite: head-to-head on text, layout, photo and comics — Negative-Whereas3307 · 2026-10-30