AssemblyAI Inc., a startup that develops artificial intelligence models optimized to process audio files, has secured $50 million in late-stage funding to support its growth efforts. 

The company announced the Series C investment today. Lead investor Accel was joined by Insight Partners, Salesforce Inc. co-Chief Executive Keith Block, former GitHub CEO Nat Friedman and a number of other backers. The round brings AssemblyAI’s total outside funding to $115 million, most of which was raised since the start of the year.

Founded in 2017, AssemblyAI offers cloud-based AI models optimized to process audio files such as call center recordings. The company’s models can transcribe speech, associate each part of the transcript with the relevant speaker and organize the text into chapters. 

AssemblyAI also helps developers analyze the contents of their audio files. The company’s AI models can extract items of interest from transcripts, such as mentions of a competitor, and generate automated summaries. Additionally, it offers a neural network optimized to detect and remove sensitive data such as customers’ credit card numbers. 

AssemblyAI debuted its newest AI model, Conformer-2, in July. It was trained on about 1.1 million hours of voice data. According to the company, Conformer-2 processes audio files nearly 50% more accurately than its previous-generation neural networks. 

Alongside cloud-based AI models, AssemblyAI provides a development framework called LeMURE. Usually, integrating a neural network into an application requires software teams to use several different development tools, which complicates their work. AssemblyAI says software teams can replace those tools with LeMURE to streamline the development process.

Several thousand organizations currently use the company’s models to process their applications’ voice data. “We’re now regularly serving over 25 million inference calls, and processing over 10 terabytes of voice data, every day through our API for our customers,” AssemblyAI founder and CEO Dylan Fox detailed in a blog post. “And, with 10,000-plus new organizations signing up for our API every month, we’re just scratching the surface of the new voice-powered AI applications we’ll see enter the market over the next year.”

The company will use the proceeds from its new funding round to hire more employees and accelerate customer acquisition initiatives. Additionally, AssemblyAI plans to procure more compute infrastructure optimized for AI training. The company will use the extra infrastructure to expand its portfolio of neural networks.

AssemblyAI is currently building a new flagship AI model, dubbed Universal, that is expected to outperform its latest Conformer-2 system in several areas. The company is training Universal on a dataset about 10 times larger than the one it used to build Conformer-2. The training is being carried out on Google Cloud instances based on the search giant’s internally-developed TPU chips, which are optimized for AI workloads.

AssemblyAI began developing Universal about six months ago. Fox detailed today that the company expects to make the model available to customers in the “very near future.”

Image: AssemblyAI