Google AI MusicLM Generates Music in Any Genre
Google has created an impressive new AI system that can generate music from a text description. But don’t be too enthusiastic about it because the tech giant has no plans in making it available publicly because of ethical concerns.
This is not the first AI system that can generate music. However, it is the first system that can create songs with high fidelity.
Thousands of Dataset
The system has been trained on a dataset of 280,000 hours of music. It can build on existing melodies, whether they are sung, played on an instrument, or whistled. Plus, it can take written descriptions and convert them into a musical story.
Then, you can direct it to create music by simply combining a picture and a caption. It can also produce music that is played by an instrument in a certain style.
Unfortunately, the results are not ideal. The samples are somehow distorted even though the system can synthesize vocals.
One of the issues that Google may face with the system is that the result may contain copyrighted materials.
During an experiment, the researchers found that one percent of the produced music was copied directly from songs it was trained. The percentage is significant so Google decides not to release it in its current form yet.
The researchers will continue to hone it and resolve the hazards associated with music generation.
The quality of the autogenerated vocals is excellent. The sounds are real.
But there’s a problem with copyrighted materials.
AI-generated music is facing legal concerns. For instance, Jay-Z’s company filed complaints against Vocal Synthesis after its channel on YouTube used AI to produce renditions such as We Didn’t Start the Fire by Billy Joel.
The films were removed. However, the video-sharing platform decided to put them back online because the takedown requests were incomplete.
Meanwhile, Google is already preparing to release its tools to compete with ChatGPT. One of them is LaMDA. It has been working on it for some time time. This is the company’s most advanced conversational AI. It was revealed in 2021 with the warning that it was still in the development phase.
Google may also release MUM or Multitask Unified Model. It lets people ask questions naturally across various types of information. This AI can answer your question that the existing search engine cannot.
The company announced how it would add it to Lens so users could take a photo of an object. For instance, you can capture a broken part of your car and get instructions on how to repair it.
ChatGPT could injure Google’s ad business model. The reason for this is that people might not need to click on links if they could just get the answer.
However, the upcoming competitor of ChatGPT might not stop Google from linking out to content.
Google emphasized that AI models could make something up. Of course, not all people would want to converse with a search engine. However, Google admitted that the conversational approach is what most users want.
The company has the technology to compete with ChatGPT and other Al tools. However, its real challenge is how to move it into actual products.