AssemblyAI Raises $50 Mn for Superhuman Speech Recognition

AssemblyAI, the company that focuses on “applied AI,” has raised $50 million in Series C funding, led by Accel, for building advanced speech recognition models for AI applications. The San Francisco-based startup specialises in the development of cutting-edge speech recognition technology. 

Apart from Accel, which also led its Series A round, other investors include Keith Block and Smith Point Capital, Insight Partners, Daniel Gross and Nat Friedman, and Y Combinator. This fund brings the company’s total funds to $115 million.

The CEO of the company, Dylon Fox, has expressed his pride in the progress made over the last two years. He believes that there is still a lot of work to be done, and the new capital will support their ambitious research plans, new model development, training compute, and market expansion. 

He said in an interview that the company aims to expand its head count and aims to build ‘Stripe for AI models.’

The blog highlights that AssemblyAI has been working on its next-gen Universal model, which is focused on multilingual speech AI tasks. Compared to Conformer-2, this model is trained on more than 10 million hours of voice data and leverages Google’s tensor processing chips (TPUs), representing a 10 times increase in training data, and 1,250 times when compared to the first model from 2019.

Conformer-2 has already set new industry standards for accuracy and robustness in various tasks such as speech-to-text and speaker identification. This model was trained on 1.1 million hours of voice data, which has resulted in up to 43% fewer errors on noisy data compared to other models. 

Highlighting the capabilities of generative AI and LLMs which the company used in its Audio Intelligence models such as Auto Chapters, Content Moderation, and its latest product LeMUR, AssemblyAI believes that the introduction of its new models will enable Speech AI capabilities even further. 

The company is now serving 25 million inference calls, while processing over 10 terabytes of voice data every day through its API customers, which include Fireflies.ai, TypeForm, Close, LoopMedia, and CallRail. 

Focusing on assisting business and enterprise use cases, AssemblyAI's technology is built on state-of-the-art deep learning models that can accurately transcribe and analyse speech data in real time for automating their voice-based workflows, enhancing customer service, and improving the accessibility of their products and services. The technology can be employed to transcribe audio and video content, generate captions for videos, and even analyse sentiment and emotion in speech data.

Moreover, the company claims that it is already being utilised by a diverse range of businesses and organisations, including media companies, call centres, and healthcare providers, which is around a 200% growth from last year.