Tech giant Google has launched an AI called MusicLM that generates music from text input.
We all have seen the power of ChatGPT – a language model that can perform a wide range of natural processing tasks, such as answering to questions, generating code from text inputs, writing poems, and many other tasks.
Now, with MusicLM, Google has leveled up the power of AI as it can generate music from the text.
What is Google’s MusicLM AI?
If we define MusicLM in one line – It is a text-to-music generation system.
The users just have to feed the appropriate text prompts into it and the music will be ready.
After getting the text from the user, MusicLM analyses it and then deciphers it to the appropriate scale and complexity to form the composition.
It is not just limited to text inputs, in fact, it can also capture whistles and hummed melodies to generate the music.
One can instruct it to generate music from a combination of text, pictures, and captions. Also, can ask it to compose specific parts of the music with different instruments and so on.
The text-to-music generation MusicLM AI is trained on 2,80,000 hours of music dataset.
MusicLM is not the first Text To Music Generation AI
While MusicLM right now can generate the most complex compositions, it is not the first AI that can transform text inputs into music.
Open AI and Infact Google already have launched text-to-music generation AI in past. For example, Open AI’s Jukebox, Riffusion, and Google’s AudioLM. But, having a limited dataset these AI can’t generate complex compositions in comparison to Google’s MusicLM.
Google researchers say MusicLM outperforms the earlier AIs in several aspects, such as generating music at 24 kHz that is consistent for several minutes plus the quality of music is also awesome.
Google hasn’t made the AI public yet
While Google has created the MusicLM AI but hasn’t made it publically available due to several ethical challenges.
One ethical challenge is: The AI is trained also on music datasets that are part of copyright material. So, the music it produces from the text inputs replicates the songs on which it is trained.
So, if Google made this AI public right now, users can replicate several existing songs without properly attributing the right owner to it.
Google researchers say “Right now, we are training the AI on existing music datasets and finding a way to make it capable of generating unique music on its own”.
What are your thoughts about Google’s MusicLM which is a text-to-music generation AI? Let me know in the comments down below.
