FireRedTTS-1S: An Upgraded Streamable Foundation Text-to-Speech System

👉🏻 FireRedTTS-1S Paper 👈🏻

👉🏻 FireRedTTS-1S Demos 👈🏻

News

• [2025/05/26] 🔥 We add flow-mathing decoder and update the technical report
• [2025/03/25] 🔥 We release the technical report and project page

Roadmap

• 2025/04
  • Release the pre-trained checkpoints and inference code.

Usage

Clone and install

• Clone the repo

https://github.com/FireRedTeam/FireRedTTS.git
cd FireRedTTS

• Create conda env

# step1.create env
conda create --name redtts python=3.10

# stpe2.install torch （pytorch should match the cuda-version on your machine）
# CUDA 11.8
conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 pytorch-cuda=11.8 -c pytorch -c nvidia
# CUDA 12.1
conda install pytorch==2.3.1 torchvision==0.18.1 torchaudio==2.3.1 pytorch-cuda=12.1 -c pytorch -c nvidia

# step3.install fireredtts form source
cd fireredtts
pip install -e . 

# step4.install other requirements
pip install -r requirements.txt

Download models

Download the required model files from Model_Lists and place them in the folder pretrained_models

Basic Usage

import os
import torchaudio![alt text](image.png)
from fireredtts.fireredtts import FireRedTTS

# acoustic llm decoder
tts = FireRedTTS(
        config_path="configs/config_24k.json",
        pretrained_path=<pretrained_models_dir>,
  )


"""
# flow matching decoder
tts = FireRedTTS(
        config_path="configs/config_24k_flow.json",
        pretrained_path=<pretrained_models_dir>,
)
"""

#same language
# For the test-hard evaluation, we enabled the use_tn=True configuration setting.
rec_wavs = tts.synthesize(
  prompt_wav="examples/prompt_1.wav",
  prompt_text="对，所以说你现在的话，这个账单的话，你既然说能处理，那你就想办法处理掉。",
  text="小红书，是中国大陆的网络购物和社交平台，成立于二零一三年六月。",
  lang="zh",
  use_tn=True
)




rec_wavs = rec_wavs.detach().cpu()
out_wav_path = os.path.join("./example.wav")
torchaudio.save(out_wav_path, rec_wavs, 24000)

Tips

• The reference audio should not be too long or too short; a duration of 3 to 10 seconds is recommended.
• The reference audio should be smooth and natural, and the accompanying text must be accurate to enhance the stability and naturalness of the synthesized audio.

⚠️ Usage Disclaimer ❗️❗️❗️❗️❗️❗️

• The project incorporates zero-shot voice cloning functionality; Please note that this capability is intended solely for academic research purposes.
• DO NOT use this model for ANY illegal activities❗️❗️❗️❗️❗️❗️
• The developers assume no liability for any misuse of this model.
• If you identify any instances of abuse, misuse, or fraudulent activities related to this project, please report them to our team immediately.