Google researchers introduced a new AI system that generates music from text descriptions.  The new model, called MusicLM, outperformed other AI music generators in both audio quality and its ability to adhere to its prompts. Due to copyright issues, Google will not release MusicLM publicly yet. More: MusicLM was trained on over 280,000 hours of music, making it able to produce tracks of varying genres, instruments, and concepts. Users type in a natural language description of what music they'd like, including the number of instruments, and the system outputs high-fidelity audio. Users can also sing, whistle, or hum to build upon a melody. The model "generates music at 24kHz that remains consistent over several minutes," according to researchers. The system is based on Google's Transformer neural network architecture, which has been particularly useful in the AI fields of natural language processing and computer vision. Google researchers also published an AI training dataset of 5,500 pieces of music to support others researchers working on automated song generation.