Open TTS Leaderboard
Hugging Face's objective-metric leaderboard for comparing multilingual text-to-speech and voice-cloning models at scale.
Overview
Hugging Face published the Open TTS Leaderboard, a Space (hf-audio/open_tts_leaderboard) that evaluates text-to-speech and voice-cloning models using objective metrics instead of slow human-preference arenas. It scores models on intelligibility via word/character error rate using Qwen3 ASR, speed via real-time factor (RTFx) and time-to-first-audio on both H200 GPU and CPU, and speaker similarity via WavLM cosine similarity, cutting evaluation time from weeks to hours. The app supports picking a language or dataset, plays a reference voice alongside generated audio from many models, and includes a 'Listen' tab for manual voting plus a streaming-performance tab. Initial results on the Seed TTS Eval and CV3 Eval benchmark splits show Kokoro-82M, Supertone/supertonic-3, and fishaudio/s2-pro leading on English WER.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Objective scoring: WER/CER via Qwen3 ASR, RTFx and time-to-first-audio on H200 GPU and CPU, WavLM speaker-similarity
- Multilingual comparisons across language/dataset selections
- Manual 'Listen' tab for human voting alongside automated metrics
- Streaming-performance tab for latency-sensitive comparisons
- Evaluation scripts planned for open-sourcing
Use cases
- Choosing a TTS or voice-cloning model for a given language and latency budget
- Benchmarking new open-source TTS models against established baselines
- Comparing speaker-similarity fidelity across voice-cloning systems
Pricing
Free, hosted as a public Hugging Face Space.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- Not listed
- Record updated
- 3 Oct 2026
- Record last checked
- 3 Oct 2026