This week, Google researchers published an article describing the results of a artificial intelligence (AI) tool designed to create music.
The tool, called MusicLM, is not the first AI music tool to be launched. But the examples provided by Google demonstrate an ability to create music based on a limited set of descriptive words.
AI shows how complex computer systems have been trained to behave in human ways.
Tools like ChatGPT can quickly produce or generate written documents that compare well to human work. ChatGPT and similar systems require powerful computers to run complex machine learning models. San Francisco-based OpenAI launched ChatGPT late last year.
Developers train these systems on massive amounts of data to learn methods for recreating different forms of content. For example, computer-generated content may include written materials, design elements, art, or music.
ChatGPT has recently received a lot of attention for its ability to generate complex writing and other content from a simple natural language description.
Google engineers explain the MusicLM system as follows:
First, a user offers a word or words describing the type of music they want the tool to create.
For example, a user might enter this short phrase into the system: “a continuous soothing violin backed by a soft guitar sound”. Descriptions entered may include different music fashionsexisting instruments or other sounds.
Several different musical examples produced by MusicLM have been published online. Some of the music generated came from descriptions of only one or two words, such as “jazz”, “rock” or “techno”. The system created other examples from more detailed descriptions containing whole sentences.
In one example, Google searchers include these instructions to MusicLM: “The main soundtrack of a arcade Game. It’s fast and optimistic, with a catchy electric guitar riff. The music is repetitive and easy to remember, but with unexpected sounds…”
In the resulting recording, the music seems to stay very close to the description. The team said the more detailed the description, the better the system can attempt to produce it.
The MusicLM model works similar to the machine learning systems used by ChatGPT. Such tools can produce human-like results because they are trained on huge amounts of data. Many different materials are introduced into the systems to allow them to learn complex skills to create realistic works.
As well as generating new music from written descriptions, the team said the system can also create samples based on a person’s own singing, buzzingwhistle or play an instrument.
The researchers said the tool “produces high-quality music…for several minutes, while being faithful to the text conditioning signal.
At present, the Google team has not released the MusicLM models for public use. This differs from ChatGPT, which was made available online for users to experience in November.
However, Google said it was releasing a “high-quality dataset” of more than 5,500 songwriting pairs prepared by professional musicians called MusicCaps. The researchers took this step to help in the development of other AI music generators.
MusicLM researchers said they believe they’ve designed a new tool to help anyone create high-quality music selections quickly and easily. However, the team said it also recognizes some risks with the machine learning process.
One of the main problems identified by the researchers was “biases present in the training data”. A bias can include too much on one side and not enough on the other. The researchers said this raises a question “about relevance for generating music for underrepresented cultures in training data.
The team said it plans to continue investigating any results from the system that might be considered. cultural appropriation. The goal would be to limit bias through more development and testing.
In addition, the researchers said they plan to continue improving the system to include speech generation, text conditioning, and better voice and music quality.
I am Brian Lynn.
Bryan Lynn wrote this story for VOA Learning English, based on reports from Google.
words in this story
artificial intelligence – nm the development of computer systems that have the ability to perform work that normally requires human intelligence
style -not. a particular shape or design, usually used to compare forms of art or handicraft
instruction -not. a description of how to do something
arcade – nm an area containing many electronic games and other coin-operated games
optimistic – adj. full of hope and happiness
repetitive – adj. say or do something over and over again
um – v. make a musical sound without opening your mouth
whistle – v. make a high-pitched sound by forcing air through a small hole in the mouth
faithful – adj. stand firm on an idea or belief
appropriate – adj. the degree to which something is suitable for a situation
cultural appropriation – nm when members of one culture in a society, often the main culture, use a practice from another, often minority culture, without fully understanding the meaning or significance of the practice.
What do you think of this story? We want to hear from you. We have a new comment system. Here’s how it works:
- Write your comment in the box.
- Below the box, you can see four images for social media accounts. They are for Disqus, Facebook, Twitter and Google.
- Click on an image and a box appears. Enter your social media account ID. Or you can create one on the Disqus system. It’s the blue circle with “D” on it. It’s free.
Every time you come back to comment on the Learning English site, you can use your account and see your comments and replies. Our feedback policy is here.