US Intelligence-Backed VC Firm In-Q-Tel Invests in Speech Tech Startup Deepgram
Speech recognition technology startup Deepgram has picked up an undisclosed strategic investment from In-Q-Tel (IQT), the venture capital arm of the U.S. intelligence community. In-Q-Tel and its government partners will now start testing Deepgram’s speech recognition and transcription technology.
Deepgram’s automatic speech transcription and analysis platform is built with a focus on enterprise services. Businesses can use the AI for collecting and organizing phone calls or meetings, similar to other transcription services. What sets the five-year-old startup’s approach to speech recognition uses end-to-end deep learning, instead of heuristics. Deepgram claims that its tech beats its rivals when it comes to accuracy, even out-of-the-box, while still performing quickly and without raising costs. That precision makes speech recognition and transcription more valuable in almost any enterprise context. The federal government is no exception, but the investment didn’t happen overnight.
“It’s an interesting dating story,” Deepgram CEO Scott Stephenson told Voicebot in an interview. “We’ve known people there for a long time and have known their investment strategy around speech. It didn’t make sense then for this arrangement, but that’s changed. A benefit of [a strategic investment is] reducing the hurdles of working with the federal government so we can provide access to Deepgram.”
Deepgram isn’t In-Q-Tel’s first dip in investing in speech technology. In 2005, the VC firm invested in CallMiner and its speech analysis software for phone calls, followed in 2009 by a deal to invest in speech recognition tech startup Carnegie Speech. In-Q-Tel’s portfolio history notes plenty of investment in related technology, such as handwriting recognition and instant translation software. The breadth of technology In-Q-Tel has been interested in makes sense considering its roots in the CIA and the fact that the Q in the name refers to the ever-inventive provider of high-tech tools for James Bond. The deal with Deepgram doesn’t mean secret agents will use the startup’s tech to transcribe conversations among spies, though, at least not yet. First, the tech will undergo some tests to see how it might fit into the government.
“At this stage, it’s more about a technology vetting,” Stephenson said. “But, wanting to test it out means it will be a fairly deep integration. It’s not just a cursory test. One key thesis of In-Q-Tel is not to morph the company into what the government needs but instead to ask, what are you building that the government can use? In the past, speech recognition was very rigid. That [Deepgram’s platform] can adapt to models is a key feature to them.”
Deepgram Digs In
The new funding deal with In-Q-Tel continues a very busy year for Deepgram. In March, the startup raised $12 million Series A funding round led by Wing VC along with NVIDIA, Y Combinator, and other investors. The investment sextupled Deepgram’s total funding to $13.9 million. Then, when the COVID-19 health crisis worsened in April, the company donated $1 million worth of its platform to assist medical providers, up to $50,000 of tech each depending on their size and need.
“We’ve had a good uptake on medical [services] since then,” Stephenson said. “I don’t think it will fade. [Speech recognition and transcription tech is] not only valuable now or if a pandemic happens again, but forever.
Automating medical transcription and analysis had started becoming more popular in the industry even before the pandemic. Medical providers could try out Amazon Transcribe Medical or the platform that Nuance and Microsoft set up to improve Nuance’s Dragon Medical Virtual Assistant. Deepgram hasn’t limited its platform to one industry in its enhancements, either. As part of the previous investment announcement, the startup revealed that it had added real-time streaming as an option, enabling the AI to transcribe and analyze conversations as they happen. The comcpany also added an on-premise deployment option. It’s the same as the standard software, except that all of the processing happens within a client’s servers, without sending any data to the cloud. Though Stephenson said the new features are unrelated to the deal with In-Q-Tel, he acknowledged that they would be beneficial to a governmental agency. They likely at least partly explain one reason for In-Q-Tel’s interest in investing in and using Deepgram’s technology.
“Deepgram’s use of an AI-enabled, neural network architecture leveraging custom speech recognition models trained on vast amounts of audio data allows them to rapidly achieve much more accurate transcriptions for non-standard audio environments vs solutions like Google Voice and Apple Siri”, In-Q-Tel managing partner George Hoyem said in a statement. “Using state of the art transfer learning also enables Deepgram to quickly build speech to text capabilities for new and novel language variants on relatively small amounts of training data, resulting in huge time savings for our government partners.”