Miso One
Miso One is the product-facing name for Miso Labs' Miso TTS 8B model, an open-weights English text-to-speech system focused on expressive and conversational speech generation. The model leverages 8 billion parameters to produce emotionally varied, natural-sounding speech with attention to rhythm, pauses, and tone. It supports prompted generation from audio context, enabling one-shot voice continuation and cloning when used with consented audio. Miso One is designed for low-latency inference, with Miso Labs reporting a latency as low as 110 ms, making it a candidate for interactive voice-agent applications. The model is accessible via a hosted demo for quick evaluation and can be deployed locally by developers willing to manage the significant GPU hardware requirements. The open-weights release includes the full model checkpoint and inference code on GitHub and Hugging Face, allowing researchers and engineers to inspect, modify, and integrate the model into their own workflows, subject to the provided license and safety guidelines.
- Category
- AI Text-to-Speech
- Pricing
- Unknown
Problem solved
Developers, ML engineers, and researchers seeking a high-quality, expressive English text-to-speech model for conversational AI face a gap between lightweight open-source TTS solutions that lack emotional depth and proprietary services with limited control. Miso One solves this by providing an open-weights 8B model that delivers realistic, emotionally varied speech with low latency, enabling local deployment, fine-tuning, and integration into voice agents without vendor lock-in.
Use cases
- Evaluate voice quality for human-like conversational AI agents
- Run local TTS experiments with a large open-weights model
- Test one-shot voice cloning using prompted audio context
- Benchmark low-latency voice-agent response times
- Inspect model architecture and training code for research purposes
- Integrate expressive TTS into a self-hosted voice application
Audience
- Developers
- ML Engineers
- Data Scientists
- Researchers
Pros
- Open weights allow full local control and customization
- Expressive, emotionally varied conversational speech output
- Reported ultra-low latency suitable for voice agents
- Supports prompted generation for voice continuation
- Public repository and Hugging Face access for easy download
Cons
- English-only generation at present
- Large 8B parameter size requires substantial GPU resources
- Early release; extended safety and watermarking practices still developing
Official website: https://miso-one.com/
Canonical LambdaBase page: https://www.lambdabase.com/tools/miso-one