TTS Arena: Benchmarking Text-to-Speech Models in the Wild

Hugging Face Blog · 2 years ago

A new benchmarking platform called TTS Arena launched to let users compare text-to-speech models by listening to samples and voting on which sounds more natural. The leaderboard includes six models at launch—ElevenLabs, MetaVoice, OpenVoice, Pheme, WhisperSpeech, and XTTS—ranked using an Elo rating system similar to chess rankings. The platform addresses the field's lack of reliable quality measurement methods by crowdsourcing human judgments instead of relying on objective metrics like word error rate or small-scale opinion surveys.