Skip to content

Latest commit

 

History

24 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

UDDETTS: Unifying Discrete and Dimensional Emotions for Controllable Emotional Text-to-Speech

Demo page; Paper; Hugging Face;

Highlight🔥

UDDETTS has been released! Compared to the previous version, some modules have been updated, and it is trained on over ten thousand hours of speech data, making the model more stable and robust.

Install

  • Clone the repo:

    git clone --recursive https://github.com/xuan3986/UDDETTS.git
    cd UDDETTS
    git submodule update --init --recursive
  • Create Conda env:

    conda create -n UDDETTS -y python=3.8.20
    conda activate UDDETTS
    pip install -r requirements.txt
  • Model download:

To ensure anonymity, pre-trained models trained on large-scale emotional speech datasets will be released on the open-source platform after review. Thank you for your patience.

Usage

[1] Download English emotional speech datasets:

  1. MSP-Podcast
  2. IEMOCAP
  3. CMU-MOSEI
  4. Expresso
  5. MELD
  6. EmoTale
  7. EU-Emotion
  8. ESD
  9. CREMA-D
  10. EmoV-DB
  11. MEAD
  12. RAVDESS

[2] Preprocess, since each dataset follows a different organization format, we handle them individually. We provide partial code and processed data samples as references.

[3] Extract features, including ADV bins, speaker embedding, speech tokens, parquet list... (Stage 1-5)

[4] Train (Stage 7)

    cd examples/exp1
    conda activate UDDETTS
    bash run.sh

[5] Inference (Stage 6)

A D V in range(1, 14), you can use a text from the test_examples.

    cd examples/exp1
    conda activate UDDETTS
    bash run.sh

Roadmap

  • 2025/09

    • We modify the code and upload some lightweight model files and data samples for demonstration.
  • 2025/05

    • Release the core architecture and base code of UDDETTS

License

The UDDETTS model can be used for non-commercial purposes, see CC BY-NC-SA 4.0. The source code in this GitHub repository is released under the following license.

Acknowledge

  1. CosyVoice.
  2. whisper.
  3. Matcha.
  4. roberta-base.
  5. 3D-Speaker
  6. hifi-gan.

Disclaimer

The content provided above is for academic purposes only. Some content is sourced from the internet. If any content infringes on your rights, please contact us to request its removal.

About

The first LLM that unifies discrete and dimensional emotions for controllable emotional TTS

Topics

Resources

Stars

10 stars

Watchers

2 watching

Forks

Packages

Contributors

Languages