The latest innovation in artificial intelligence
For its latest innovations in the field of artificial intelligence,
Google revealed a music generation performance capable of converting a short text command
And the year 2022 witnessed a major revolution in the field of artificial intelligence, especially when OpenAI launched the ChatGPT chatbot and the artificial intelligence image creation platform DALL-E 2, and this coincided with the emergence of other advanced systems of artificial intelligence with capabilities to create images or texts according to requests and instructions. users
However, not many people were as enthusiastic about these musical tools as they were about text or image creation tools, due to the fact that their results are very ordinary; Most of them are of low quality and fidelity, and lack the basic structure of the song.
However, there have been many improvements to these tools, and perhaps the most prominent of them is the MusicLM tool from Google, which creates music using artificial intelligence; It is able to create 5-minute audio clips in proportion to the entered text, and it is true that these audio clips are not of high quality, but they are more like human songs than to music created using other artificial intelligence systems.
Google used more than 280,000 hours of music recordings to train the MusicLM tool, using MuLan, a model trained to associate music with descriptions written in natural language.
The company then created MusicCaps, a database of more than 5,500 publicly available music recordings for the purpose of evaluating the quality of the tracks produced by the AI system.
Music experts put captions for each clip, adding a description of each one and categorizing them according to genre, for example.
Google then compared MusicLM with two artificial intelligence systems capable of converting text into music, namely Mubert and Riffusion
and for this purpose adopted a set of quantitative measures to assess the sound quality of the music clip, and its adherence to the text.
Google Rating Experts; MusicCaps database captions and two music tracks; They could both be created by AI, or one could be created by AI
and the other is from the music from which the database took its description and annotation. Google asked the experts to pick out which tracks of music best match the given descriptions
Google researchers published their results in a scientific paper, noting that MusicLM is superior to other systems in all fields.
In general, MusicLM may be able to produce music close to the music created by humans, but it is still unable to simulate the structure of traditional songs, especially since the quality of the music it produces is very low, and its lyrics are incomprehensible.
However, Google believes that future improvements will focus on addressing these issues, improving
the overall quality of audio, and eliminating the problem that prevents this tool from being releas
“We acknowledge the potential intellectual misappropriation risks of creative content associated with this issue, and we stress the need for future improvements to head off the risks associated with AI music generation,” the researchers wrote.
/written commands/artificial intelligence/google/music