Google’s New AI Musiclm Can Turn Text Into Music
In the last few months, ai technology just become booming throughout cyberspace, we know few of them such as ChatGPT (and we already reviewed it), Dall-e and now we have MusicLM. Google researchers just made an ai that can turn text-prompt into minutes-long musical instrument, and even can turn a whistled into any other instrument. It’s similar to Dall-e that can turn text prompts into images, this MusicLM turn text prompts into music. Sadly, while we still can’t play around with it, the company has uploaded a bunch of samples that it produced.
The samples are no kidding, it’s impressive. There a lot of samples from 30 seconds piece of music that generated from a paragraph long description that describe a genre, what it used for, vibe, even a specific instrument. There’s also a 5 minutes song from a basic text prompt such as relaxing jazz. There’s also a story mode generate music where the audio is generated from a sequence of text prompts. Not only that there’s also text and melody conditioning, a painting caption conditioning, a short 10 second audio generated from instrument, genre, musical experiences, even places and epochs. You can listen the generated audio here.
Of course, it’s not for everyone to enjoy, but still, it’s produced a good instrumental result according to the text prompts. MusicLM even can emulate basic tone of human vocals even though is far from good but it’s convincing enough, and of course by no means it’s perfect, some results sound strange and vocals tends to be inaudible. This result was produced when the AI asked to make music that would be played at the gym. If you’re hearing it yourself, you’ll find that the audio was so raw that it was inaudible and the lyrics are nonsense but like we said before, it was convincing enough.
This is not the first time AI is used in a music generator, before this we have like Google’s AudioML, OpenAI’s Jukebox that have the similar functions. However, MusicLM’s model have a vast training database almost 280,000 hours of music to help it learn and produce a coherent song with better variety and depths.
The picture above that we get from the research pepper is explaining it in detail of how is “hierarchical sequence-to-sequence modeling task” that the researchers use along with AudioLM, another Google project.
Google said in their research paper that to support future research they released a dataset composed of 5.5k music-text pair with rich text descriptions provided by human expert called MusicCaps. As of now, there is no word on when this model will be made available to the public. The reason the researchers aren’t releasing MusicLM to public is because of copyright concern, there’s roughly 1% of the music that was produced at the time of publications was copied directly from the training data.
All in all, it was both highly impressive and equally terrifying what AI can do, especially lately with the AI trending. This model alone, MusicLM could provide fuel for fire for those who argue that AI has the potential to destroy music productions as we know it.
Google appears to be feeling the pressure from latest AI trending especially OpenAI’s ChatGPT. Due to its ability to give customers the information they need in an approachable style, artificial intelligence chatbots have taken the tech industry by storm in recent months. According to The New York Times, Google has altered its plans in response to ChatGPT's threat to its search business over the past few weeks.
The report also claimed that CEO Sundar Pichai declared a “Code Red” and accelerated AI development, google even reportedly preparing at least 20 AI powered products and chatbot for its search engine this year with at least some of it will be shown at Google I/O conference in May. Pichai also reported brought in Google founders Larry Page and Sergen Brin last month to meet with current leaders to review AI plans and offer input. According to reports, Google has made an effort to speed up the product clearance process, including tests to make sure AI-driven technology is fair and ethical. The corporation is also reportedly changing the amount of risk it is willing to accept as it introduces such technology. The search chatbot experiment appears to prioritize accuracy, screening out false information, and safety. The Times reported that Google has " a lower bar and will try to curb issues relating to hate and toxicity, danger and misinformation rather than preventing them," for the other goods and capabilities it is developing.
It’s not that difficult why Google is said to be in panic over this AI trend especially AI powered chat bot such as ChatGPT. One of its reasons is earlier this month that reports suggested that Microsoft as an investor of OpenAi plans to incorporated some of the tech that powering ChatGPT into Bing. The company also said that ChatGPT will soon be incorporated into the Azure OpenAI Service.
With all AI trending lately, it’s important to keep everything in checks. Always stay informed, double check every information you get especially from AI. Use AI as a tool to help you acquiring new skills, learn new things, not the other way around making ai do every creative activity.
picture from electronicbeats.ne
