Published on

VALL-E X: Multilingual Text-to-Speech Synthesis and Voice Cloning

  • avatar

VALL-E X: Multilingual Text-to-Speech Synthesis and Voice Cloning 🔊

Are you fascinated by the capabilities of text-to-speech (TTS) technology? Have you ever wondered how it would feel to have a multilingual TTS model that can clone voices and synthesize speech with different emotions? Look no further! Introducing VALL-E X, an open-source implementation of Microsoft's groundbreaking zero-shot TTS model.


VALL-E X is an amazing multilingual text-to-speech (TTS) model proposed by Microsoft. While Microsoft initially published their research paper on this model, they did not release any code or pretrained models. Recognizing the potential and value of this technology, our team took on the challenge to reproduce the results and train our own model. We are glad to share our trained VALL-E X model with the community, allowing everyone to experience the power of next-generation TTS! 🎧

The VALL-E X model is capable of multilingual speech synthesis in English, Chinese, and Japanese. It can generate natural and expressive speech in these languages, making it a versatile tool for various applications. Additionally, VALL-E X supports zero-shot voice cloning, which means you can enroll a short 3~10 seconds recording of an unseen speaker and watch the model create personalized, high-quality speech that sounds just like them!

How Does It Work?

VALL-E X is built on the concept of GPT-style audio generation. It predicts audio tokens quantized by EnCodec, a powerful audio encoder-decoder. The model's computation complexity increases quadratically as the sequence length increases, so all training is kept under 22 seconds. To generate long text, a huge paragraph must be broken down into short sentences.

To use VALL-E X, you can install it with pip and Python 3.10. You will also need CUDA 11.7 ~ 12.0 and PyTorch 2.0+. Once installed, you can generate audio from text prompts using the provided Python code. The generated audio can be saved to disk or played directly in the notebook.

Benefits and Use Cases

VALL-E X comes packed with cutting-edge functionalities that make it a powerful tool for various applications:

  1. Multilingual TTS: VALL-E X can speak in three languages - English, Chinese, and Japanese - with natural and expressive speech synthesis. This makes it a valuable tool for creating content in multiple languages and reaching a wider audience.

  2. Zero-shot Voice Cloning: With VALL-E X, you can enroll a short recording of an unseen speaker and generate personalized speech that sounds just like them. This opens up possibilities for voice-over work, virtual assistants, and more.

  3. Speech Emotion Control: VALL-E X can synthesize speech with the same emotion as the acoustic prompt provided. This adds an extra layer of expressiveness to your audio and allows you to create engaging content.

  4. Zero-shot Cross-Lingual Speech Synthesis: VALL-E X can produce personalized speech in another language without compromising on fluency or accent. This feature is particularly useful for language learning apps, translation services, and more.

  5. Accent Control: VALL-E X allows you to experiment with different accents, like speaking Chinese with an English accent or vice versa. This can be a fun and creative way to engage with your audience and add a unique touch to your content.

  6. Acoustic Environment Maintenance: VALL-E X adapts to the acoustic environment of the input, making speech generation feel natural and immersive. This ensures that the generated speech sounds realistic and blends well with the surrounding audio.

Future Directions

While VALL-E X is already a powerful and versatile TTS model, there are several areas of improvement and future directions that our team is actively working on:

  • Fine-tuning for better voice adaptation: We are constantly working on improving the voice cloning capabilities of VALL-E X to make the generated speech even more personalized and accurate.

  • Replace Encodec decoder with Vocos decoder: We are exploring the possibility of using Vocos decoder, a state-of-the-art audio decoder, to further enhance the quality and realism of the generated speech.

  • Support for more languages: We are actively working on expanding the language support of VALL-E X to include more languages and cater to a wider range of users.


VALL-E X is a groundbreaking multilingual text-to-speech (TTS) model that brings a new level of expressiveness and versatility to speech synthesis. With its support for multiple languages, zero-shot voice cloning, and various other advanced features, VALL-E X is a valuable tool for content creators, developers, and AI enthusiasts alike. We are excited to share our trained model with the community and look forward to seeing the innovative applications that will be built upon it.

To learn more about VALL-E X and access the pretrained model, visit the GitHub repository. Feel free to explore the demos and try out the model's capabilities hassle-free. If you have any questions or need assistance, don't hesitate to reach out to us on our Discord channel.

Happy voice cloning! 🎤