What is LeVo
LeVo is an AI singing model launched by Tencent AI Lab, featuring powerful timbre cloning capabilities. It can accurately replicate a target timbre, including pitch, emotion, and rhythm, with just 3 seconds of audio, without requiring large amounts of training data. LeVo supports stem generation, allowing separate generation of vocal and accompaniment tracks, which facilitates post-editing. Its technical architecture is based on a language model (LM), combining LeLM and a music codec to generate tracks in parallel. Its sound quality performance is close to the industry-leading level, and it performs exceptionally well in lyric alignment capability.
Main Features of LeVo
-
Zero-shot timbre cloning: Only 3 seconds of audio are needed to accurately replicate the target timbre, including pitch, emotion, and rhythm, without requiring large amounts of training data.
-
Stem generation: LeVo supports a dual-track generation mode, allowing separate generation of vocal and accompaniment tracks, providing greater flexibility for later mixing and editing.
-
High-fidelity music performance: LeVo's sound quality performance is close to the industry-leading level, especially excelling in musicality, harmony between vocals and accompaniment, and sound quality (MOS score). Through multi-preference alignment methods to optimize generation results, it ensures that music maintains a high-fidelity effect across various styles and scenarios.
Technical Principles of LeVo
- Language model architecture: LeVo adopts a language model (LM) architecture, combining LeLM and a music codec to generate high-quality musical works in parallel.
Performance of LeVo
-
It is comparable to the industry-leading Suno4.5 on multiple key indicators.
-
In lyric alignment capability (LYC), LeVo is 0.21 points higher than Suno4.5, demonstrating excellent text control capability.
Project Address of LeVo
- Project official website：https://levo-demo.github.io/
Application Scenarios of LeVo
-
Individual music creators: Provides a low-threshold, high-quality music creation platform for individual users who love music creation but lack professional skills.
-
Professional music producers: The stem generation feature and high-fidelity music performance can improve creation efficiency and quality.
-
Music education institutions: It can be used to provide students with vivid and interesting music teaching experiences.