Meta, formerly known as Facebook, has introduced its latest language learning model (LLM) called MusicGen, designed to generate AI-generated music. Developed by Meta’s internal Audiocraft team, MusicGen functions similarly to ChatGPT but focuses on creating music based on text descriptions provided by users.
Users can enter a brief text description specifying the type of music they want to hear, click the “Generate” button, and within a short time, MusicGen produces a 12-second track based on the provided instructions. For instance, a user can request a “lofi slow BPM electro chill [song] with organic samples,” and MusicGen delivers audio reminiscent of the popular YouTube Lofi Girl radio.
To enhance the AI’s understanding of musical structure, users can “steer” MusicGen by uploading their own songs. Felix Kreuk, one of the developers behind the LLM, shared samples on his Twitter profile showcasing MusicGen’s ability to take Sebastian Bach’s renowned Toccata and Fugue in D Minor and infuse it with drum beats and 1980s-style synths, resulting in an upbeat rendition of the piece.
MusicGen is currently available to the public on Meta’s Hugging Face website, allowing anyone to try it out. However, it’s worth noting that unlike Google’s AI music generator MusicLM, Meta’s model is limited to instrumental tracks and cannot generate vocals. This limitation is perhaps for the best, as MusicLM’s vocal output resembles Simlish, a fictional language that is unintelligible.
Musicians need not be concerned about their careers being threatened by AI-generated music. While MusicGen is capable of creating simple, short melodies, it lacks the depth and quality that comes from human ingenuity. Some tracks generated by MusicGen can become repetitive as the AI cycles through the same progressions multiple times. The tool proves useful for generating background audio for videos or presentations but falls short of creating truly engaging music. The next pop hit won’t be AI-generated, at least not yet.
In conclusion, Meta’s MusicGen represents a step forward in AI-generated music, providing a convenient tool for creating simple instrumental tracks based on user descriptions. However, the current capabilities of MusicGen are limited, and the quality of its output does not match that of music created by human musicians. While it serves a purpose in certain contexts, true musical innovation and engagement still heavily rely on the creativity and expertise of human musicians.
