Like every other creative field, AI is completely changing the music industry. Today, Google DeepMind unveiled a new state-of-the-art AI system for music generation. Dubbed Lyria, the model shows unprecedented skill in creating original songs complete with lyrics, instrumental backing tracks, and vocals.
Lyria represents a massive leap forward for AI music creation. Unlike other AI music generation models that struggle with maintaining musical continuity, Lyria excels in generating high-quality music complete with instrumentals and vocals, offering users refined control over style and performance.
DeepMind credits Lyria's advances to its deep learning techniques, which allow it to better model musical coherence across long sequences. Specifically, Lyria excels at several key tasks:
- Generating high-quality vocals and instrumentals from scratch that sound cohesive over the length of an entire track. The AI accounts for continuity in lyrics, melody, rhythm, and other musical elements.
- Performing music style transfer and continuation. For example, Lyria can take an existing song and convincingly transform it into another genre or compose a seamless additional verse.
- Allowing granular user control over the output to adjust attributes like genre, mood, instruments, and more. This provides creators with nuanced direction rather than just random generation.
- Watermarking the AI-generated audio with a novel spectrogram technique. This sheds light on the track's origins while being imperceptible to listeners.
YouTube has also announced that it is collaborating with Google DeepMind to leverage Lyria to power a new experiment called Dream Track on its platform. A select group of popular artists, including Charlie Puth, John Legend, T-Pain and others, have partnered with YouTube and DeepMind to test Dream Track. This feature allows creators to generate a custom 30-second soundtrack for their YouTube Shorts using the unique vocal style and flair of the participating musicians.
Dream Track prompts users to enter a topic or theme. It then presents a carousel of the partner artists to choose from. Upon selecting an artist, Lyria generates a short original track with lyrics, instruments, and the AI-replicated voice of the chosen musician - all tailored to the user's theme.
Early samples show Lyria convincingly emulating the musical styles of different artists. The vocals mimic distinctive elements like Charlie Puth's emotive crooning or T-Pain's auto-tuned rap-singing.
YouTube states Dream Track explores AI's potential to foster deeper connections between artists, creators, and fans. However, they emphasize responsible development, working closely with the music industry to shape Lyria's deployment. The platform has even partnered with DeepMind to watermark AI-created tracks using novel audio recognition technology.
Alongside Dream Track, DeepMind unveiled collaborative work on new AI music tools for artists and producers. The tool will include features that allow transforming melodies and sounds between instruments, generating accompaniments, and more. DeepMind will continue refining the tools with feedback from industry partners.
While still in early experimental stages, Lyria and Dream Track provide a glimpse into an AI-powered future for music creation. They demonstrate the technology's promise in augmenting human artistry rather than replacing it. With responsible development, such innovations could enable breakthroughs in musical creativity, engagement, and access.