Meta, formerly known as Facebookhas recently launched a new AI-powered music generator called MusicGen. This innovative tool can transform text descriptions and melodies into audio compositions. As part of its commitment to openness and collaboration, Meta has made the MusicGen code and models publicly available for research purposes to encourage reproducibility within the music community.
Felix Kreuk, a research engineer at Meta AI research, presented MusicGen as a simple and controllable music generation model on Twitter. The tool adopts a single-stage transformerLM architecture built on top of the EnCodec audio tokenizer. This design incorporates efficient patterns of token weaving, which eliminates the need for multiple cascading models such as upsampling or hierarchical structures.
To ensure the robustness and versatility of the model, Meta trained MusicGen on a large dataset consisting of 20,000 hours of music. The dataset comprises 10,000 meticulously selected high-quality licensed music tracks as well as 390,000 instrument-only tracks sourced from ShutterStock and Pond5 media libraries.
While Meta’s MusicGen brings a new dimension to AI-generated music, it’s worth noting that other tech giants have delved into this space too. In May, Google introduced ‘MusicLM,’ an experimental AI tool that generates high-fidelity music in various genres based on text descriptions. Initially announced in January this year, MusicLM is now available to the public via web browsers or Android and iOS devices through the AI Test Kitchen app.
The emergence of AI-powered music generators like MusicGen and MusicLM underscores growing interest and potential for AI in creative domains. These tools provide musicians, songwriters and enthusiasts with powerful resources to experiment with inspiration and create new musical compositions. By making code and models freely available for researchers’ use in further advances in music generation research ultimately enriching the musical community while facilitating exploration of capabilities offered by AI creativity.
According to India TV News ( Meta and Google aim to foster collaboration and further advances in music generation research by making the code and models freely available. This will ultimately enrich the music community while facilitating exploration of the capabilities offered by AI creativity.
