MiniMax, the next-generation speech model, has topped the international authoritative speech evaluation list Speech-2 on ArtificialAnalysis.

MiniMax's Speech-2 has topped the international authoritative speech evaluation list ArtificialAnalysis, beating out OpenAI and ElevenLabs. Speech-2 uses a TTS model with a self-regressive transformer architecture, achieving zero-sample voice cloning and a new Flow-VAE architecture, improving voice generation quality and similarity. Speech-2 also has features such as super-human, personalization, and diversity, supporting 32 languages and performing well in Chinese, English and other languages. MiniMax is exploring landing solutions for scenarios such as voice assistants, voice chatting, and dubbing, and accelerating the commercialization process.