What Is MusicGen?
Meta’s MusicGen is an innovative AI tool that can generate short, original pieces of music based on text prompts, which can optionally be aligned to an existing melody. Built on a Transformer model, MusicGen operates similarly to a language model, predicting the next section in a piece of music instead of the next characters in a sentence.
To enable efficient processing, the researchers behind MusicGen utilized Meta’s EnCodec audio tokenizer to break down the audio data into smaller components. This single-stage model processes tokens in parallel, resulting in fast and efficient music generation.
The training of MusicGen involved an extensive dataset of 20,000 hours of licensed music. The researchers relied on a combination of an internal dataset containing 10,000 high-quality music tracks, as well as music data sourced from Shutterstock and Pond5.
One of MusicGen’s standout features is its ability to handle both text and music prompts. Users can input text prompts that define the desired style, which is then matched to the melody in the audio file. For example, by combining a text prompt specifying a “light and cheerful EDM track with syncopated drums, airy pads, and strong emotions, tempo: 130 BPM” with the melody of Bach’s renowned “Toccata and Fugue in D Minor (BWV 565)”, MusicGen can generate a corresponding piece of music.
It’s important to note that precise control over the orientation to the melody is limited. The text prompt provides a rough guideline for the generation process but may not be exactly reflected in the output.
In comparison to Google’s MusicLM, MusicGen holds a distinct advantage. The researchers conducted tests on three versions of their model with different sizes, and while larger models produced higher quality audio, the 1.5 billion parameter model was rated the best by humans. On the other hand, the 3.3 billion parameter model excelled in accurately matching text input to audio output.
Meta’s MusicGen represents a groundbreaking advancement in the field of music creation. By leveraging the power of AI, this tool enables the generation of unique compositions based on text and melody prompts. While legal challenges may arise, the emergence of AI-generated hits highlights the transformative potential of AI in the music industry. Musicians, marketers, and creatives of all kinds can now explore uncharted territories and reshape the future of music as we embrace this exciting new frontier.
