Miso One
Open-weights 8B AI text-to-speech model for expressive English speech.
āĻ¸ā§āĻā§āĻ°
60
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°āĻāĻžāĻ°ā§āĻ° āĻ°ā§āĻāĻŋāĻ
3.0
āĻŦāĻŋāĻ¨āĻžāĻŽā§āĻ˛ā§āĻ¯ā§/āĻā§āĻ°āĻžāĻ¯āĻŧāĻžāĻ˛ āĻ¸āĻšāĻžāĻ¯āĻŧāĻ¤āĻž
āĻ¸āĻŽāĻ°ā§āĻĨāĻŋāĻ¤ āĻ¨āĻ¯āĻŧ
āĻļā§āĻ°ā§āĻ° āĻŽā§āĻ˛ā§āĻ¯
--
āĻ¸āĻ°ā§āĻŦāĻļā§āĻˇ āĻāĻĒāĻĄā§āĻ
āĻāĻāĻ¸ā§āĻ 14, 2026
āĻ¸āĻāĻā§āĻˇāĻŋāĻĒā§āĻ¤āĻ¸āĻžāĻ°
Miso One āĻā§?
Miso One is an open-weights, 8B-parameter text-to-speech (TTS) system developed by Miso Labs. It is designed specifically for producing highly realistic, expressive, and emotionally varied English conversational speech, making it ideal for voice-agent research and developer workflows. Built on a Sesame-style conversational speech model (CSM) architecture with Mimi audio codes, it features a highly optimized inference capability boasting a published low latency of 110 ms. In addition to text-to-speech generation, the model supports voice continuation and one-shot voice cloning from audio context with clear consent boundaries.
Miso One āĻā§ āĻāĻžāĻā§ āĻŦā§āĻ¯āĻŦāĻšā§āĻ¤ āĻšāĻ¯āĻŧ?
āĻā§āĻ¯āĻžāĻ
āĻļā§āĻ°ā§āĻˇ āĻŦā§āĻļāĻŋāĻˇā§āĻā§āĻ¯
- 8B-parameter open-weights text-to-speech model
- Expressive, conversational, and emotionally varied English speech
- Low-latency architecture with a published 110 ms response claim
- One-shot voice cloning and prompt-audio continuation capabilities
- Discrete audio-code modeling using Mimi audio codes
- 48 kHz preview quality with built-in watermarking and safety checks
Miso One āĻā§āĻāĻžāĻŦā§ āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ° āĻāĻ°āĻŦā§āĻ¨?
Users can evaluate Miso One by reading its official model card on the repository or Hugging Face page, trying out the hosted web demo to check voice quality, or downloading the public 8B weights and inference code to run local benchmarks within their own CUDA environment. For hosted creator workflows, users can sign up and choose a subscription plan based on their required annual or monthly character capacity.
āĻŦāĻŋāĻāĻ˛ā§āĻĒ āĻā§āĻ˛
āĻā§āĻ¨ āĻŦāĻŋāĻāĻ˛ā§āĻĒ āĻā§āĻ˛ āĻĒāĻžāĻāĻ¯āĻŧāĻž āĻ¯āĻžāĻ¯āĻŧāĻ¨āĻŋ
āĻ¸ā§āĻŦāĻŋāĻ§āĻž & āĻ āĻ¸ā§āĻŦāĻŋāĻ§āĻž
āĻā§āĻ¨ āĻĄā§āĻāĻž āĻ¨ā§āĻ
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°ā§āĻ° āĻā§āĻˇā§āĻ¤ā§āĻ°ā§
- Running low-latency conversational speech generation locally for voice-agent research
- Generating high-quality, expressive voiceovers, narrations, and captions for creator content
- Creating instant voice clones and personal voice models using brief, consented audio context prompts
āĻŦā§āĻ¯āĻŦāĻšāĻžāĻ°āĻāĻžāĻ°ā§āĻ° āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻž
3.0
ā§Ģ āĻāĻ° āĻŽāĻ§ā§āĻ¯ā§
3.0
āĻāĻāĻ¨āĻ āĻā§āĻ¨ āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻž āĻ¨ā§āĻ
āĻāĻ āĻā§āĻ˛ā§āĻ° āĻĒā§āĻ°āĻĨāĻŽ āĻĒāĻ°ā§āĻ¯āĻžāĻ˛ā§āĻāĻ¨āĻžāĻāĻžāĻ°ā§ āĻšāĻ¨!