HeartMuLa – Open-Source Music Generation Model (github.com)

HeartMuLa has announced the release of an open-source family of music generation models that aim to revolutionize music creation through AI. This suite includes HeartMuLa itself, a versatile music language model capable of generating music based on lyrics and tags in multiple languages; HeartCodec, a high-fidelity music codec; HeartTranscriptor, designed for accurate lyrics transcription; and HeartCLAP, which aligns audio and text for improved cross-modal retrieval. Notably, the latest version, HeartMuLa-7B, showcases competitive performance in musicality and fidelity, positioning it as a leading tool for AI music generation. This release holds significant implications for the AI/ML community, especially in creative fields. By providing improved lyrics controllability and audio quality, HeartMuLa empowers musicians and content creators to experiment with AI in their work seamlessly. The introduction of comprehensive benchmarks, such as HeartMuLa-Benchmark, enhances the framework for evaluating AI-generated music, ensuring transparency and fostering further innovation. With continuous enhancements through reinforcement learning and open-source deployment, HeartMuLa sets a high standard in the intersection of AI technology and music, promoting collaboration and creativity across diverse genres and languages.