Google is being hailed for its impressive efforts after developing a new AI system that may produce music of any genre through the likes of a text description.
However, the company is not celebrating just yet as they fear the risks included are plenty. Therefore, the Android maker is not announcing any plans for a release, anytime soon.
While we agree the tech giant is certainly not the first to produce such a generative AI system for songs but it’s definitely getting a lot of praise as we speak. The company has its own list of technical limitations and restrictions on training data. But none of those were able to generate songs, especially linked to a complex composition.
But this new project called MusicLM is one of the first of its kind that could do something like this. It was stated in a paper once how it received training through the likes of about 280,000 music hours so it could learn well and produce tunes in a coherent manner via text descriptions.
The songs are so good that at times, you’ll be amazed at how close they resemble the likes of a human artist’s composition. But it might be a challenge for anyone to overstate that the samples are good. The fact that there are no musicians or even instrumentalists on board, it’s truly amazing.
The capability of MusicLM is beyond just producing short sound clips of tunes. The researchers from Google mentioned how such a system can design a whole system of just melodies that are sung, hummed, and even whistled. Moreover, you see it being played on the likes of instruments too. But wait, that is not all.
MusicLM could get instructed via the likes of picture and caption combos or even make audio that gets played by a certain type of instrument of a particular type. Moreover, even the experience of AI musicians could get set up and the system may end up creating tunes that are inspired by a few places.
But like many other options, it definitely does come with its fair share of flaws. It’s not perfect and a lot of the products generated have some huge flaws and distorted quality. This is just one example of side effects from training that can’t be avoided.
Google’s new AI system may produce vocals technically such as choral harmonies, but there is still plenty to be desired. A lot of the lyrics are ranging from the extreme of being poor English to purely gibberish. And it’s all sung by voices that appear as if several singers’ voices were amalgamated.
But despite all of this, researchers at Google mentioned that the challenges are plenty and a lot of them are linked to ethical behavior. There is a tendency to entail material that’s copyrighted from places like training data that get produced into tunes.
Meanwhile, during one certain experiment, around 1% of the music that was produced by the system was called out for replication of tunes from which it received training.
Read next: Hybrid Work Might Not Be Suitable for Introverts According to This Report
However, the company is not celebrating just yet as they fear the risks included are plenty. Therefore, the Android maker is not announcing any plans for a release, anytime soon.
While we agree the tech giant is certainly not the first to produce such a generative AI system for songs but it’s definitely getting a lot of praise as we speak. The company has its own list of technical limitations and restrictions on training data. But none of those were able to generate songs, especially linked to a complex composition.
But this new project called MusicLM is one of the first of its kind that could do something like this. It was stated in a paper once how it received training through the likes of about 280,000 music hours so it could learn well and produce tunes in a coherent manner via text descriptions.
The songs are so good that at times, you’ll be amazed at how close they resemble the likes of a human artist’s composition. But it might be a challenge for anyone to overstate that the samples are good. The fact that there are no musicians or even instrumentalists on board, it’s truly amazing.
The capability of MusicLM is beyond just producing short sound clips of tunes. The researchers from Google mentioned how such a system can design a whole system of just melodies that are sung, hummed, and even whistled. Moreover, you see it being played on the likes of instruments too. But wait, that is not all.
MusicLM could get instructed via the likes of picture and caption combos or even make audio that gets played by a certain type of instrument of a particular type. Moreover, even the experience of AI musicians could get set up and the system may end up creating tunes that are inspired by a few places.
But like many other options, it definitely does come with its fair share of flaws. It’s not perfect and a lot of the products generated have some huge flaws and distorted quality. This is just one example of side effects from training that can’t be avoided.
Google’s new AI system may produce vocals technically such as choral harmonies, but there is still plenty to be desired. A lot of the lyrics are ranging from the extreme of being poor English to purely gibberish. And it’s all sung by voices that appear as if several singers’ voices were amalgamated.
But despite all of this, researchers at Google mentioned that the challenges are plenty and a lot of them are linked to ethical behavior. There is a tendency to entail material that’s copyrighted from places like training data that get produced into tunes.
Meanwhile, during one certain experiment, around 1% of the music that was produced by the system was called out for replication of tunes from which it received training.
Read next: Hybrid Work Might Not Be Suitable for Introverts According to This Report
