Jano TTS

Jano-TTS is a lightweight, fast multi-speaker Text-to-Speech model based on Flow Matching and a Matcha-style UNet architecture. With only ~24.6M parameters, it supports 128 speakers and generates high-fidelity 24kHz audio via the Vocos vocoder.

⚡ Quick Start

1. Install Requirements

pip install -q torch transformers vocos tokenizers

2. Run Inference

from IPython.display import Audio
import torch
from transformers import AutoModel

device = "cuda" if torch.cuda.is_available() else "cpu"

# Load the model
model = AutoModel.from_pretrained(
    "jano3/jano-tts", 
    trust_remote_code=True
).to(device)

# Generate audio
audio = model.inference(
    text="Hello world! Jano TTS is running smoothly.",
    spk_id=99,            # Speaker ID (0 - 127)
    temperature=0.7,
    cfg_strength=2.0,
    n_steps=16,           # 8-16 steps (use fewer steps on CPU for faster inference)
)

# Play the output
Audio(audio, rate=24000, autoplay=True)

More info & Fine-Tuning Guide: https://github.com/laki35/jano-tts/blob/main/README.md

Citation

If you use this model in your research or project, please cite it as follows:

@misc{jano_tts,
  author = {Jano3},
  title = {Jano-TTS: Lightweight Flow-Matching Text-to-Speech},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/jano3/jano-tts}}
}