TTS benchmarks
Every benchmark the TTS index tracks, with each core benchmark's share of the composite and how many models it covers.
13 benchmarks tracked.
- Artificial Analysis Speech Arena: Controlled Voice Elo (voice cloning)
- Artificial Analysis Speech Arena: Provider Voice Elo
- TTS Arena V2 (Hugging Face, TTS-AGI) Elo
- CV3-Eval: multilingual cloning (CER/WER, SIM)
- EmergentTTS-Eval: overall win rate vs gpt-4o-mini-tts baseline
- MiniMax TTS Multilingual Test Set: speaker similarity
- MiniMax TTS Multilingual Test Set: WER (24 languages)
- Seed-TTS-eval: Mandarin CER (zero-shot cloning intelligibility)
- Seed-TTS-eval test-hard: CER
- Seed-TTS-eval: speaker similarity SIM-o (WavLM-large SV)
- Seed-TTS-eval: English WER (zero-shot cloning intelligibility)
- Time to first audio byte (TTFB), independent measurement
- TTSDS2 distribution score (prosody, speaker, intelligibility, generic)