Image Credits: CreateSafe
After a song using AI deepfakes of Drake and the Weeknd’s voices was taken down after becoming a viral hit, Grimes shocked the public when she tweeted that she would split 50% of the revenue with anyone who wanted to use her voice in AI-generated songs. A week later, Grimes debuted Elf.Tech, an open-source AI voice software that allows artists to replicate her voice in their songs and pocket half of the royalties.
Despite many musicians feeling threatened by AI, the Canadian electronic artist (and co-parent to children with Elon Musk who has just launched an AI company of his own) is doubling down on her stance regarding the use of AI music tools.
Music tech studio CreateSafe, co-founded by Grimes’ manager Daouda Leonard, officially launched today its generative AI-powered platform, Triniti — available on the web as a free open beta. Triniti enables artists to create an AI voice clone, generate text-to-audio samples, ask a chatbot music industry-related questions, monetize creations and manage music IP.
Elf.Tech is the prototype of CreateSafe’s Triniti platform. Elf.Tech has created more than 1,000 songs replicating Grimes’ voice. Grimes recently joined the CreateSafe’s advisory board.
In addition, CreateSafe revealed that it raised $4.6 million in funding, the company exclusively told TechCrunch. The seed round was led by Polychain Capital with participation from Kendrick Lamar’s manager Anthony Saleh, Paris Hilton-founded 11:11 Media, Unified Music Group, Crush Ventures, MoonPay, Chaac Ventures and Dan Weisman, the vice president of Bernstein Private Wealth Management. CreateSafe will use the new capital to further develop the Triniti platform.
CreateSafe refers to Triniti as an “artistic intelligence” platform since its main purpose is to empower creators to make music.
“It’s not going to be fully AI because you have this interface to play around with and make the music on your own,” Leonard said during a demo with TechCrunch. He has worked with many top artists, including DJ Snake and Skrillex. “That’s why we call it artistic intelligence because we want everyone to own their AI and be able to collaborate and scale themselves in a new way.”
Triniti leverages a customized version of an RVC (Realistic Voice Cloning) model, as well as OpenAI’s ChatGPT for its virtual companion and Stable Diffusion to generate images for album covers. CreateSafe is also building a music model based on MusicGen that focuses on audio techniques like traditional DSP (Digital Sound Processing).
“Our goal is not to replace humans with computers. It’s to give humans computers that collaborate with them to create a new experience around the creation of music, assisted by AI… We don’t want to contribute to the massive amount of fraud that’s happening with music now where people are uploading music just to game the system,” Leonard added.
For context, it is estimated that at least 10% of streaming activity was fraudulent in 2022, according to Beatdapp, a company that specializes in audit and fraud detection.
The voice transformation and cloning feature is the most notable AI tool that Triniti offers. It allows singers to record their voices and train the AI with different voice patterns and styles. Artists are then given a digital voice clone along with a “voice likeness” certificate, allowing them to set terms on how their voice is licensed. Songs created using the voice can be distributed, managed and monetized directly from the platform.
The licensing part is key here since a lot of debates around voice AI in the music industry are centered on copyright protections.
“Giving the licensing framework and the distribution pathway for songs is a way to protect your voice when establishing ownership over this digital clone… If another platform is potentially using your voice likeness with your name, you can issue a takedown,” Leonard said.
Even though Grimes is the first artist to use Triniti’s voice cloning technology, she won’t be the last. The platform has a cohort of 30 artists who plan to release digital voice clones in 2024 for Triniti members to utilize for their own work. Leonard declined to share who; however, he said they are “well-known, legacy artists.”
“We wanted to start off with a cohort of artists and record labels who really believe in this technology… We showcase things like the ethical use of how you train a model and the consensual use of getting permission from the IPO holder,” Leonard explained.
Another tool worth touching on is Triniti’s AI sample generator, where creators input a genre or vibe that they want the audio to sound like. For instance, “catchy yet haunting love song.” The generator creates four tracks to choose from, each paired with album images. Meanwhile, if an artist needs advice, Triniti has a virtual companion that can answer questions such as, “How do labels make money?” or “How do I get famous?”
Lastly, Triniti’s management tool uses LLMs to automate workflow, helping managers catalog deals, create invoices, add new clients and organize projects.
As of today, anyone can access Grimes’ AI voice tech, audio samples and chat for free. Voice cloning, licensing, distribution and management apps are available upon request/approval. The company eventually plans to introduce a subscription model, ranging from $99 to $150 per year.
In the near future, Triniti will introduce editing tools, a MIDI (Musical Instrument Digital Interface processing visual and more. It will also launch iOS and Android apps.
“I’m really proud of the team that they were able to execute this so fast and with such vision,” Grimes said in a statement to TechCrunch. “There’s a lot to talk about, but ultimately, art generates so much money as an industry, and artists see so little of it. A lot of people talk about abundance as one of the main end goals of tech, acceleration, AI, etc. For us, the first step is actually figuring out how to remove friction from the process of getting resources into artists’ hands.”