Miso One

Description

Miso One is the product-facing site for Miso TTS 8B, an 8B-parameter, open-weights text-to-speech model from Miso Labs. It is built for expressive, conversational English speech, voice continuation, and low-latency voice-agent research, and Miso One helps creators, developers, and researchers understand and evaluate the model before deploying it.

The site offers a free online AI voice generator to try realistic speech, with a hosted demo for judging voice quality, emotional range, and English pronunciation. The current public release is English-only and uses discrete audio-code modeling; Miso Labs publishes a 110 ms low-latency claim intended to be benchmarked in your own environment. Prompted generation from audio context supports one-shot voice continuation, which should be evaluated with consented audio only.

Miso One points to the official Hugging Face model page and MisoTTS GitHub repository for local inference, and provides model facts, usage history, safety and watermarking notes, and pricing. Voice plans use credits shared across TTS, Voice Design, and Voice Clone, with free users limited to 120 characters per conversion and paid plans and credit packs allowing up to 1,000 characters.

Miso One's Core Features

• Free online AI voice generator and hosted demo

• Based on Miso TTS 8B open-weights model

• Expressive, conversational English speech

• Low-latency generation for voice-agent research

• Prompted generation for one-shot voice continuation

• Links to Hugging Face weights and GitHub for local inference

• Model facts, safety, and watermarking notes

• Credit-based plans shared across TTS, Voice Design, and Voice Clone

How to use Miso One?

1. 1

   Read the model card: Start with the official Hugging Face page and MisoTTS repo to confirm license, checkpoint status, and requirements.

2. 2

   Try the hosted demo: Use the demo to judge voice quality, emotional range, and English pronunciation.

3. 3

   Run local inference: Install the repo, download the 8B weights, and benchmark latency and memory in your own environment.

4. 4

   Test prompt audio: Evaluate voice continuation with consented audio prompts before deciding whether it fits your project.

Miso One's Use Cases

• Voice-agent research
• Model evaluation
• Local TTS deployment
• Voice continuation

FAQ from Miso One

• What is Miso One?

  It is an information and demo site for Miso TTS 8B, an open-weights English text-to-speech model from Miso Labs for expressive, conversational speech.

• Is Miso One multilingual?

  No. The current public release is focused on English only and should not be described as a broad multilingual product.

• How large is the model?

  The current public model is an 8B-parameter text-to-speech release, Miso TTS 8B.

• Can I run it locally?

  Yes. The public files on the MisoTTS GitHub repo and Hugging Face are aimed at developers who can run and evaluate the model in their own CUDA environment.

• How much can I generate for free?

  Free users can generate up to 120 characters per conversion, while paid plans and credit packs allow up to 1,000 characters.

• Does it support voice cloning?

  The model supports prompted generation from audio context for voice continuation, which should be evaluated with consented audio only.