Phonic AI Platform Receives Investment from Lux Capital

The Evolution of AI Voice Technology and Phonic's Innovative Approach

The current standard of quality in artificially generated voices is sufficient for applications such as audiobook creation, podcasting, text-to-speech functionality, and fundamental customer service interactions. However, a significant number of businesses remain hesitant to fully implement this technology due to concerns regarding its overall dependability.

Introducing Phonic: A Comprehensive Voice Solution

Recognizing this gap in the market, Moin Nadeem and Nikhil Murthy, both graduates of MIT (pictured above), established Phonic. Their company is dedicated to providing a complete voice stack designed to enhance the reliability of synthetic voices while simultaneously minimizing latency.

Nadeem and Murthy’s acquaintance extends back over seven years, originating during their time together at MIT. The development of Phonic commenced last year, driven by their observation that few companies were concentrating on delivering fully integrated voice technology solutions.

The Need for Scalable Reliability

“The current state of Voice AI often involves combining separate components like automatic speech recognition and text-to-speech, followed by the integration of intelligence,” explained Murthy in an interview with TechCrunch. “However, our conversations with potential customers revealed a distinct shortage of solutions that demonstrate consistent reliability when scaled for larger operations.”

Nadeem, formerly employed at MosaicML – a company acquired by Databricks for $1.3 billion in 2023 – noted that many companies operating within the voice AI landscape, such as Vapi and Rounded, are focused on assembling workflows using disparate AI models.

Phonic's End-to-End Model Training

Phonic distinguishes itself through a different strategy: it develops and trains its models internally, encompassing the entire process from start to finish. Murthy highlighted several benefits associated with this approach.

“Maintaining ownership of the models enables us to integrate crucial reliability features directly into the models themselves,” he stated. “Without control over this foundational layer, you are essentially connecting independent elements that may not function together seamlessly.”

Furthermore, Murthy emphasized that Phonic’s methodology facilitates efficient model hosting and execution, leading to cost savings. The company trains its models using a diverse range of audio recordings, including those with accents and background noise, to ensure exceptional robustness.

Current Partnerships and Future Availability

Currently, Phonic is collaborating with a select group of partners in sectors like insurance and healthcare. A broader product launch is anticipated within the coming months. Prospective clients will soon have the opportunity to evaluate Phonic’s technology directly through the company’s website, as Nadeem indicated.

Seed Funding and Investor Confidence

Phonic has successfully secured $4 million in seed funding, led by Lux, with additional participation from prominent figures including Amjad Masad (co-founder of Replit), Clem Delangue (co-founder of Hugging Face), Qasar Younis (co-founder of Applied Intuition), and Erik Bernhardsson (founder of Modal Labs).

Lux Capital's Perspective

Grace Isford, a partner at Lux Capital, expressed the firm’s enthusiasm for Phonic’s in-house model training approach.

“We hold both Moin and Nikhil in high regard as exceptional technologists,” she commented. “Their founding of a machine learning club at MIT and their extensive experience in model training were particularly compelling. Additionally, their innovative combination of diffusion and proprietary models within the voice AI sector is truly noteworthy.”