Enterprise Speech Recognition Startup Deepgram Raises $12M
Speech recognition technology startup Deepgram has closed a $12 million Series A funding round led by Wing VC along with NVIDIA, Y Combinator, and other investors. The funding comes as the company debuts new features for its enterprise-focused automatic speech transcription and analysis platform.
Businesses are growing to rely on artificial intelligence to record, transcribe, and analyze phone calls, meetings, and other audio. The speed and low cost of using AI can make up for the errors and sometimes imprecise analysis that results. Deepgram claims its approach can outperform the industry standard without sacrificing the speed and cost benefits.
“The old way of doing speech recognition is not going to be the same as future versions,” Deepgram CEO Scott Stephenson told Voicebot in an interview. “The heuristics model of speech recognition has too many flaws; it’s why I wanted to start Deepgram. When we began, we decided to start from scratch with end-to-end deep learning. The AI now can learn their jargon, learn the background noises at their meetings, and learn their natural cadence of speaking.”
Almost five years after its founding, Deepgram’s platform now offers multiple price points depending on just how precise the client wants the transcription and analysis to be. Stephenson explained that because Deepgram is built on deep learning rather than heuristics, adjusting precision is relatively straightforward, and the standard version platform can outperform most other options.
“No one else is operating on the scale we offer with our level of accuracy,” Stephenson said. “Generally, with our competitors, you’re talking 70% accuracy, but with some of our customers, we reach over 90% accuracy. More accuracy means allowing your model to be exposed and do what a human would do when they come to work for you in terms of paying attention. The customers can pay a little more to make it more accurate, but a lot of times, the out-of-the-box accuracy is already higher than they are used to, so people run with it. Then, six months or a year later, they want a custom model.”
New Money, New Features
The $12 million funding round marks a steep increase in investment for Deepgram. The company has picked up $13.9 million in funding, all told. The money will go to iterating the platform and boosting Deepgram’s client list, Stephenson said. Two of those new features were announced along with the funding. Clients concerned about privacy but who want Deepgram’s services can now get on-premise deployment. The platform operates as usual, but it all happens within a client’s servers, without sending any data to the cloud. Deepgram also started offering real-time streaming to its clients, which transcribes and analyzes conversations as they are happening. Speeding up that process while maintaining accuracy is something a lot of companies are very keen on, Stephenson said.
“Understanding is an even bigger concern than speech recognition,” Stephenson said. “It’s a messy and persistent problem for enterprises that we can solve. We built something that would find those exciting spots in meetings. Now, it can find them faster.”
The infusion of capital is well-timed since Deepgram is far from alone on the field. IT packages by giants like Microsoft, Google, and Amazon usually include automatic transcription and enterprise tech mainstays Salesforce and Cisco are developing or integrating better AI for audio analysis. Deepgram isn’t the only transcription AI startup either. There are others raising capital to compete in the space, such as the $10 million raised by Otter in January and the $5 million that Fireflies picked up in December.