Google develops MusicLM, a revolutionary AI music generator
The system was trained on a dataset of 280,000 hours of music, allowing it to generate songs of complexity
Google has developed a cutting-edge AI system, known as MusicLM, that is capable of generating music in any genre given a text description. However, the company has no immediate plans to release it due to concerns about the potential risks.
Although MusicLM is not the first generative AI system for song, it stands out from previous attempts such as Riffusion, an AI that composes music by visualising it, Dance Diffusion, Google’s own AudioML and OpenAI’s Jukebox.
These earlier systems were limited by technical constraints and a lack of sufficient training data, resulting in songs that were less complex in composition or lower quality. MusicLM, however, has overcome these limitations and can produce highly intricate and high-fidelity songs.
MusicLM was trained on a dataset of 280,000 hours of music, allowing it to generate songs from descriptions. The generated songs have a human-like quality, though they may not be as original or musically cohesive.
The samples produced by MusicLM are noteworthy for their high quality, considering there are no musicians or instrumentalists involved in the process. Even when provided with complex descriptions, MusicLM is able to capture instrumental riffs, melodies and moods.
Additionally, the AI’s experience level can be set and it can create music inspired by different places, epochs, or purposes such as motivational music for workouts. However, MusicLM is not without its flaws, some of the samples produced have a distorted quality and the vocals generated are not of high quality. The lyrics are often gibberish and the singing is done by synthesised voices that sound like a combination of several artists.
