Deepgram raises $25M to help enterprises tap into speech recognition, voice data potential
Deepgram, which provides technology for voice biometric and speech recognition solutions, has announced a $25 million Series B funding round led by Tiger Global with participation from Citi Ventures, Wing VC, SAP.io, and NVIDIA Inception GPU Ventures.
The new funds will now enable Deepgram to continue development efforts of its speech recognition technology in key market differentiators such as accuracy, speed, scale, and affordability.
The San Francisco-based company will also use the new funds to improve its technology further, grow its team, and continue delivering highly accurate transcriptions.
Last year was an eventful one for Deepgram, with the company closing a successful Series A round, hiring 47 new employees, and reporting a total of 100 billion spoken words processed.
Also in 2020, the firm introduced a new training feature called Deepgram AutoML, to further improve the development of artificial intelligence-powered models.
According to Deepgram, 90 percent of all data in the world is unstructured, including audio, images, and video. The firm’s goal is to exploit this opportunity, in regards to voice data in particular, in the hope of gaining more insight related to businesses, customers, and markets.
“Voice is the largest treasure trove of enterprise data waiting to be unlocked,” said Zach DeWitt, Partner at Wing VC. “Deepgram is modernizing speech recognition for the modern enterprise, making it fast and simple to unearth valuable data from any conversation, meeting, or customer interaction.”
According to the executive, today’s speech recognition technology is designed for simple queries, such as the one used by smartphone voice assistants, but enterprise speech is more complex.
“You have multiple people speaking over each other with industry-specific jargon,” DeWitt explains. “Deepgram’s end-to-end deep learning approach is revolutionizing what data can be extracted from voice, providing enterprise customers unparalleled accuracy and scalability.”
Sharpen VP of Product Adam Settle said Deepgram was able to build custom voice recognition models, making it a clear choice to partner with.
The company trains its speech models to learn and adapt to complex, real-world scenarios. These take into consideration customers’ unique vocabularies, accents, product names, and background noise.
Despite all these variables, however, Deepgram reports an average transcription accuracy of over 95 percent.
AI | biometrics | Deepgram | enterprise | funding | research and development | speech recognition | voice biometrics | voice recognition