QinYue Large Music Model 173

Qinle Large Model is an AI music large model jointly developed by Tencent Qin Lab and Tencent AI Lab. It features rich AI composition and arrangement abilities. Simply input Chinese or English keywords, descriptive sentences, or audio, and Qinle Large Model can directly generate music.

Main Features of Qinle Large Model:

• Music Generation: Qinle Large Model can intelligently generate music based on user-provided Chinese or English keywords, descriptive sentences, or audio input, enabling automatic music creation.

• Score Generation: In addition to generating audio, Qinle Large Model can also generate detailed musical scores containing multiple tracks such as melody, chords, accompaniment, and percussion, providing users with rich musical structures.

• Automatic Editing: The model supports a series of automatic editing operations on the generated scores, including score continuation, regenerating specific tracks or measures, adjusting instrumentation, and modifying instrument types and rhythms.

• Audio-Text Alignment: Through contrastive learning techniques, Qinle Large Model constructs a shared feature space that aligns audio tags or text descriptions with the audio itself, providing conditional control signals for the generation model and enhancing the relevance and accuracy of music generation.

• Score/Audio Representation Extraction: The model can convert musical scores or audio into a series of discrete feature (token) sequences, which provide the foundation for large language model predictions.

• Large Language Model Prediction: Using a decoder-only architecture, Qinle Large Model is trained through feature prediction (next token prediction). The predicted sequences can be converted back into scores or audio, achieving the conversion from text to music.

• Audio Restoration: Through flow matching and vocoder techniques, Qinle Large Model can restore the predicted audio representation sequences into audible audio, enhancing the realism and quality of the audio.

• Music Theory Compliance: During the music generation process, Qinle Large Model follows music theory, ensuring that elements such as melody, chords, and rhythm conform to musical logic and human aesthetics.

Application Areas of Qinle Large Model

• Music Creation: Provides professional musicians with efficient composition and arrangement tools, simplifying tedious tasks in the creative process.
  

• Education Field: Helps students understand and learn music theory, and practice through automatically generated musical scores.

• Entertainment Industry: Ordinary users can also experience the charm of music creation through simple operations, increasing interactivity and entertainment.

Tencent Music Qin Lab focuses on multimedia and AI application technology research and implementation for products such as QQ Music, Quanmin K Song, Tencent Music People, Kuwo, and Kugou. The star projects it is responsible for include music recognition, humming recognition, cover recognition, lyrics timestamp, premium sound quality, DMEE, MV video editing, subtitle recognition, and celebrity recognition on QQ Music; singing scoring, intelligent pitch correction, intelligent sound effects, and singing voice synthesis on Quanmin K Song; AI reading and AI manuscripts on Lazy Listen; intelligent quality inspection, highlight detection, and end-cloud integrated video quality enhancement in live interactive streaming. At the same time, it is also exploring virtual human technology in the metaverse. The virtual humans of Xiaotian and Xiaoqin siblings, centered on virtual images, virtual dance, singing expression generation, and music light shows, are about to be unveiled. Stay tuned.

Currently, the related technologies of Qinle Large Model have been launched on the Tencent Music Qiming Star platform, where users can register and experience it for free.