Hear Clearly: Voice Isolation for Noisy Environments
▼ Summary
– Subtle Computing develops voice isolation technology to capture clear speech in noisy environments, addressing a key challenge for voice AI applications.
– The startup creates device-specific models that preserve acoustic characteristics, achieving significantly better performance than generic voice solutions.
– Founded by Stanford graduates, the company has raised $6 million in seed funding from notable investors including Entrada Ventures and founders of Twitter, Pinterest, and Perplexity.
– Qualcomm selected Subtle Computing for its voice and music extension program, ensuring compatibility with Qualcomm chips in OEM devices.
– The company plans to launch a consumer hardware/software product next year while currently partnering with unnamed consumer hardware and automotive brands.
The challenge of capturing clear speech in noisy settings is a major hurdle for the growing number of voice-activated AI tools and services. A California startup named Subtle Computing is tackling this issue head-on with its specialized voice isolation technology, designed to ensure devices understand users perfectly, whether in a bustling cafe or a busy office.
The market for consumer applications using voice AI is expanding rapidly. Popular AI meeting assistants such as Granola, Fireflies, Fathom, and Read AI are attracting significant user adoption and investor interest. Established firms like OpenAI, ClickUp, and Notion have incorporated voice transcription features, while newer app developers like Wispr Flow and Willow are advancing voice dictation capabilities. On the hardware side, companies including Plaud and Sandbar are creating dedicated devices to record speech, using AI to generate insights and enable interaction.
A persistent difficulty for all these products is reliably picking up a user’s voice in less-than-ideal acoustic conditions. To solve this, Subtle Computing built an end-to-end voice isolation model that can discern speech even with significant background noise. According to co-founder Tyler Chen, while many companies focus on voice comprehension, the common practice of sending audio to the cloud for cleaning is inefficient.
The startup’s unique approach involves training specific models tailored to the acoustic properties of individual devices and adapting to each user’s voice. This stands in contrast to creating a single, generic model intended to function across all hardware. Chen explained that by preserving a device’s specific acoustic characteristics, they achieve performance that is an order of magnitude better than one-size-fits-all solutions. This methodology also enables them to provide personalized voice solutions to the end-user.
The founding team, Tyler Chen, David Harrison, Savannah Cofer, and Jackie Yang, met during their time at Stanford University. Chen, Cofer, and Yang were PhD candidates, while Harrison was completing his MBA. They collaborated in Steve Blank’s Lean Launchpad course, initially exploring alternative computing interfaces, which eventually led them to establish Subtle Computing.
Chen envisions a future where conversational interaction with our devices becomes the norm. He points out that the critical question is how well these devices can comprehend us in our everyday environments. Whether someone is in an extremely loud coffee shop or a shared office space discussing private matters, current voice technology often falls short.
The startup’s model for voice isolation is compact, requiring just a few megabytes of storage, and operates with a latency of only 100 milliseconds on certain devices. For other hardware, the company can deploy a separate model that both isolates the voice and provides a text transcription. Chen emphasized that their isolation technology significantly improves the accuracy of their transcription model, leading to more reliable text output.
Subtle Computing’s progress has been recognized by Qualcomm, which selected the startup for its voice and music extension program. This partnership ensures the startup’s technology will be compatible with Qualcomm’s chipsets and available on devices manufactured by original equipment manufacturers (OEMs).
To support its development, the company secured $6 million in a seed funding round led by Entrada Ventures. The round included participation from Amplify Partners, Abstract Ventures, and a group of angel investors featuring prominent 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 a former director of an early project at Alphabet’s X, observed that the voice AI sector is crowded, yet the actual user experience is frequently subpar. She believes Subtle Computing’s dedicated focus on voice isolation offers a fresh and valuable perspective to the market. Davis stated that while opinions may vary on AI’s impact on daily life, advancements in computing power and machine learning present clear opportunities for breakthroughs in voice interfaces, if executed properly. She described Subtle Computing as a game-changer for delivering reliable, easy, and enjoyable voice experiences that perform consistently in both extremely noisy and very quiet settings.
The startup has also formed partnerships with an unnamed consumer hardware brand and an automotive brand to integrate its solutions. However, Subtle Computing has ambitions beyond merely supplying models to other businesses. The company has announced plans to launch its own consumer product, which will integrate both hardware and software, slated for release next year.
(Source: TechCrunch)
