Voice Isolation Models | Noise Reduction - Subtle Computing

Subtle Computing Advances Voice Isolation Technology

A California-based startup, Subtle Computing, is actively developing innovative voice-isolation models to address the challenges of capturing clear audio in challenging, noisy environments. This technology holds significant potential for enhancing the performance of voice-driven AI products and services.

Growth in Voice AI Applications

The adoption of voice AI within consumer applications is experiencing substantial growth. Companies such as Granola, Fireflies, Fathom, and Read AI, specializing in AI-powered meeting note-taking, have garnered considerable attention from both users and investors.

Established tech firms like OpenAI, ClickUp, and Notion have already integrated voice transcription capabilities into their platforms. Furthermore, developers at Wispr Flow and Willow are concentrating on voice dictation technologies.

Hardware manufacturers, including Plaud and Sandbar, are leveraging dedicated devices to transcribe speech, subsequently utilizing AI for data analysis and interactive experiences.

The Challenge of Noisy Environments

A primary obstacle for these companies lies in accurately capturing user voices across diverse settings, including bustling cafes and busy office spaces.

Subtle Computing’s Approach

To overcome this hurdle, Subtle Computing has engineered an end-to-end voice-isolation model capable of discerning speech even amidst significant background noise. According to Chen, numerous organizations are engaged in voice understanding research.

He highlighted that a common practice involves transmitting audio to the cloud for processing, which isn’t always the most efficient solution.

Device-Specific Model Training

The startup’s strategy centers on training specialized models tailored to the acoustic properties of individual devices and adapting to each user’s unique vocal characteristics. This contrasts with the approach of developing a single, universal model.

“By preserving the acoustic fingerprint of a device, we achieve performance improvements of an order of magnitude compared to generic solutions,” Chen explained. “This also enables us to deliver personalized experiences to each user.”

Founding and Origins

Subtle Computing was founded by Tyler Chen, David Harrison, Savannah Cofer, and Jackie Yang, who initially connected while at Stanford University. Chen, Cofer, and Yang were PhD candidates, while Harrison pursued an MBA.

Their collaboration began within Steve Blank’s Lean Launchpad course, where they explored alternative computing interfaces and ultimately established Subtle Computing.

The Future of Voice Interaction

“As our interactions with AI evolve, we are moving towards a future where voice communication with our devices becomes commonplace,” Chen stated. “However, a critical question arises: how well do our devices truly understand us, the users, in the various environments where we operate daily?”

He continued, “Whether it’s a crowded coffee shop or a shared office where private conversations may occur, current voice technology often falls short.”

Model Efficiency and Performance

The startup reports that its voice-isolation model can operate on certain devices with a footprint of just a few megabytes and a latency of 100 milliseconds.

Additionally, the company offers a separate model for transcribing speech and generating text output for other devices. Chen emphasized that the isolation model enhances the accuracy of the transcription model, resulting in more reliable transcripts.

Qualcomm Partnership

Subtle Computing has been selected as a member of Qualcomm’s voice and music extension program, ensuring compatibility with Qualcomm’s chipsets and availability on devices manufactured by original equipment manufacturers (OEMs).

Seed Funding

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

Industry Perspective

Karen Roter Davis, managing partner at Entrada Ventures and a former director at X (Alphabet), observed that the voice AI landscape is competitive. While voice interactions are increasing, the overall user experience remains suboptimal.

She believes that Subtle Computing’s focus on voice isolation offers a unique and valuable perspective. “Advances in computing power and machine learning present opportunities for breakthroughs in voice interfaces – when executed effectively,” Davis said.

“Subtle Computing is addressing user needs with voice interfaces that perform reliably in both noisy and quiet environments, creating a voice experience that is intuitive, enjoyable, and transformative,” she added.

Strategic Partnerships and Future Plans

Subtle Computing has established partnerships with a consumer hardware brand and an automotive manufacturer, though the specific companies remain undisclosed, to integrate its solutions.

However, the startup’s ambition extends beyond simply supplying models to other companies. Subtle Computing intends to unveil a consumer product encompassing both hardware and software in the coming year, with further details to be announced.