Subtle Computing’s voice isolation models help computers understand you in noisy environments

California-based startup Subtle computing is tackling the problem of capturing people’s voices in noisy environments with its own voice isolation models – a technology that could benefit voice-based AI products and services.

Consumer apps that use voice AI are experiencing tremendous growth these days. AI meeting note takers such as Granola, Fireflies, Fathom, and Read AI have gained the attention of users and investors alike. Existing companies such as OpenAI, ClickUp and Notion have integrated voice transcription solutions. App makers such as Wispr Flow and Willow are working on voice dictation. Then there are hardware companies like Plaud and Sandbar that use devices as a medium to transcribe your voice and then use AI to generate insights and interaction.

One of the challenges for these companies is capturing the voices of users in any environment, such as noisy cafes or offices.

To address this, Subtle Computing has developed an end-to-end voice isolation model that can understand what you say, even in noisy environments. Chen said there are many companies working on voice understanding. He noted that device manufacturers sometimes send the voice to the cloud to get a clean output, but that is not efficient.

The startup trains specific models that match the acoustics of a particular device and adapt to the user’s voice, rather than training one model that works across devices.

“What we found is that when we preserve the acoustic properties of a device, we get an order of magnitude better performance than generic solutions. This also means that we can provide personalized solutions to the user,” said Chen.

The company was founded by Tyler Chen, David Harrison, Savannah Cofer and Jackie Yang, who met at Stanford. Chen, Cofer and Yang were pursuing their PhDs while Harrison was pursuing an MBA. They came together in Steve Blank’s Lean Launchpad course, where they worked on alternative interfaces for computers and started building Subtle Computing.

“As we interact more with AI, we are moving towards a future where we talk to our devices,” Chen said. “But the obvious question is to what extent our devices understand us, the users, in all the environments we work in every day. Whether it’s a super loud coffee shop or a shared office where other people are around you and you might be talking about something private – voice doesn’t work that way anymore these days,” he added.

The startup said it can run the voice isolation-only model on some devices, which is only a few megabytes in size and has 100ms latency. The company may also use a different model to transcribe the voice and provide text output for other devices. Chen said the isolation model allows the company’s transcription model to better understand users, and in turn creates a more accurate transcript.

Subtle Computing said Qualcomm has selected the startup as a member of its voice and music expansion program. This means the startup’s technology would be compatible with Qualcomm’s chips and available on devices produced by OEMs.

The company has raised $6 million in seed funding led by Entrada Ventures, with participation from Amplify Partners, Abstract Ventures and angel investors, including founders such as Twitter’s Biz Stone, Pinterest’s Evan Sharp and Perplexity’s Johnny Ho.

Karen Roter Davis, managing partner at Entrada Ventures and former director of an early project at X (Alphabet), noted that voice AI is a noisy space, and while interactions through this medium are increasing, the overall voice experience isn’t great. She thinks the startup’s focus on voice isolation brings a different perspective to the market.

“While you can debate whether AI will increase or decrease daily time commitments, we can all agree that advances in computing power and machine learning/AI provide opportunities for breakthroughs in voice interfaces – if done right,” said Davis. “Subtle Computing is meeting people where they are with voice interfaces that hold up in extreme noise and extremely quiet, delivering a voice experience that is reliable, convenient and fun. It’s a game changer,” she added.

The company said it is also working with a consumer hardware brand and an automotive brand — without naming them — to implement its solutions. But Subtle Computing doesn’t just want to be a model supplier for other companies.

The startup also said it plans to announce a consumer product next year that includes both hardware and software, without providing details.