Xiaomi MiMo-V2.5-TTS Open API: supports singing, natural language emotional modulation, and audio cloning of voice in seconds

ME News Report, April 24 (UTC+8), according to Beating Monitoring, Xiaomi released the MiMo-V2.5-TTS series of speech synthesis models, providing services through the MiMo open platform API, with free limited-time testing during the public beta. The series includes three models, each targeting different scenarios. MiMo-V2.5-TTS has built-in high-quality voices, supports singing mode, and can accurately express pitch and rhythm. MiMo-V2.5-TTS-VoiceDesign supports generating entirely new voices from a natural language description, without reference audio, allowing definitions based on age, gender, accent, temperament, and other dimensions. MiMo-V2.5-TTS-VoiceClone clones voices by providing a few seconds of reference audio to replicate the target speaker’s voice, preserving breath, rhythm, and pauses, without training or fine-tuning. All three models support controlling speech style via natural language commands, such as adjusting emotion with descriptions like “gentle but tired” or “tender in agitation,” and also support precise control with audio tags (e.g., “inhale,” “laughter,” “choking”). Language support includes Chinese, English, as well as dialects like Northeastern, Sichuan, Henan, Cantonese, and others. Audio output sampling rate is 24,000 Hz, with streaming output recommended in PCM16 format. (Source: BlockBeats)