Jul 10, 2023 432

Voice.ai raises $6M as its real-time voice changer approaches 500K users

#Artificial Intelligence

Services like Midjourney and ChatGPT have pushed the boundaries of how AI can create images and text out of basic text prompts. Now, audio appears to be the inevitable next frontier. Music generation based on word prompts, AI tutors for language learning and voice simulators have all seen developments in recent months. Voice.ai hopes to be a part of that conversation (heh) with technology that lets users change (and disguise) their voices in real time, and now it has raised its first outside funding on the heels of early growth.

With more than 480,000 users and a library of more than 50,000 voice filters, Voice.ai has picked up $6 million, funding that it plans to use to take its voice changing tech into new places. Mucker Capital and M13 are leading the round. Before now, Voice.ai has grown by word of mouth — the startup has a Discord channel with more than 120,000 people — on the back of $3 million in self-funding.

Currently the company’s tools — available as apps for Mac, PC, Android and iOS — are getting adopted by gamers, content creators, Vtubers and others on TikTok, Zoom, Discord, Minecraft, GTA5, Fortnite, Valorant, League of Legends, Among Us, Skype, WhatsApp and other platforms. The Voice.ai interface lets them create a new voice, or select from some 50,000 different pre-created voices (created and shared by users like themselves), which can be used as-is or modified, to use live in supported platforms, or for recordings. The plan is to use the funding to hire more technical talent and to build new SDKs and APIs to work with further platforms like Meta, Unreal and Unity; bring on multi-language support; and add in new applications like singing where voice is center stage. The startup doesn’t single it out, but it will be interesting to see if it uses some of the funding also to increase server capacity.

That is no small burden. Anecdotally, we’ve heard that GPU pain is one of the biggest gating factors in how a lot of AI apps are able to scale at the moment. (It’s partly why you’re seeing big deals being made that include strategics providing processing and server capacity.) For Voice.ai specifically, your voice is processed locally and channeled into wherever it will be used through what founder and CEO Heath Ahrens described to me as a “virtual audio cable.” But when you look at reviews of its apps, a common lament is that when you sign up you are put on a waitlist because “overwhelming demand has our servers at max capacity” with a promise that you’ll be informed when the service increases that capacity.

There are dozens of speech-to-voice and voice-to-speech services in the market today, and already a lot of activity among them: Last year Spotify acquired Sonantic and Snap bought an AI voice assistant even earlier than that; another startup, Sanas, is working on changing your accent and there are the voice simulators Murf and Acapela, among many others. Voice.ai counts itself in the same general category as Respeecher and ElevenLabs, two voice-to-voice AI startups, letting users apply masks to tweak or completely transform their voices — in some cases creating completely synthetic voices in place of the real thing. Respeecher, founded and based in Ukraine, made a name for itself by helping build a new Darth Vader voice for new Star Wars installments, based on how James Earl Jones sounded 45 years ago when he originated the role. (In keeping with a character hell-bent on destroying worlds, Darth’s voice was delivered to the Hollywood client from its offices in Ukraine as Russia marched into the country.) ElevenLabs — famously (or infamously as the case may be) — has built a platform that is frighteningly good at cloning voices, and earlier this month it picked up its most recent funding round of $19 million from a group of big-name investors. Voice.ai is trying, in that mix, to position itself as the AI voice modifying app for Everyman.

“There are plenty of companies that are trying to provide a different flavor of voice tech to businesses,” Ahrens told TechCrunch in an email (ironically, it wasn’t possible to arrange a live interview with him). Ahrens has some experience with the building of B2B AI tech: his two previous companies — iSpeech for text-to-speech and Haystack for face recognition — are built around API offerings. “What sets Voice.ai apart is that we are focused on bringing tech that was previously reserved for enterprise companies directly into the hands of consumers in an affordable fashion.” Many users, he noted, “come to us from classical DSP voice changers and voice modulators which they had been using in the past and which are still popular among many gamers and streamers.” “Affordable” comes in two tiers, with most users now on a free service that requires them to opt in to providing computational power to train Voice.ai’s models, with its service built on its own private data set comprised of “millions of unique users.” No pricing is provided on the site: we’re asking for those details. “We believe in making technology accessible and plan on working together with the open source community to democratize Voice AI technology,” added Ahrens.

Voice.ai also claims it takes what is a fundamentally different approach to the challenge of changing a voice, tapping into some of the ethos that has built up around the use of avatars by Vtubers, gamers and others online. “Most voice AI companies that are coming into the space try to build scalable enterprise focused text-to-speech solutions or expensive voice-to-voice services for production studios,” Ahrens said. “We start from the opposite spectrum and try to deliver value to individuals who are looking to expand how they sound online. The core value proposition of our speech-to-speech AI isn’t that it can perfectly replicate any given person. It’s that it retains the core elements of a user’s speech: their emotion, pacing and emphasis while replacing the sound of the voice, in order to create a completely unique new end result, in real-time.” It might be because of how the demographics in interactive platforms like gaming skew, but for now Voice.ai’s audience is 70% male versus 30% female with new categories opening not just around who is using the tech, but why. That includes not just those using avatars and building voices to match them, or those looking for more privacy protection, but also, he said, “transgender users who can represent themselves with voices that match their identity, as well as users exploring completely new online personas for themselves.”

There is already a base of users tapping into Voice.ai’s direct-to-consumer offerings, but one of the reasons why Mucker is investing in the startup is because it believes that there is an opportunity to build out a network of developers using and integrating its tech. “Voice.ai is poised to revolutionize the AI developer community in a manner akin to AdMob’s impact on the mobile app developer community,” said Omar Hamoui, a partner at lead investor Mucker Capital. (Hamoui previously founded the mobile ad startup AdMob, eventually acquired by Google, so he has some direct experience building mobile developer tools.) “By offering user-friendly solutions that were once exclusive to large enterprises, Voice.ai aims to democratize access for developers worldwide.”

Karl Alomar, the former COO of Digital Ocean, who led the investment for M13, said investors will be taking an active role in the next stage of development. “At Digital Ocean too we saw the value of building a community of builders by builders,” he said. “We’re excited for creators and developers to build on the Voice.ai platform.”