AI speech platform Deepgram has raised $47 million in Series B funding led by Madrona. In total, the company raised $72 million with this transaction, making it the biggest known Series B round in the automatic speech recognition category.
"We're building the foundational AI models for understanding human speech. Understanding speech at scale starts with accurate transcriptions, but it doesn't stop there. At Deepgram, we view accurate transcription as an increasingly solved problem for many of the more than 100 languages we work with. That's the groundwork upon which we're building the future of speech understanding: to give our customers insight into not just what was said, but how it was said, which can result in an actionable understanding of why it was said, " said Scott Stephenson, Co-Founder and CEO of Deepgram.
Unlike the traditional, text-based approach with Natural Language Processing, Deepgram leverages audio-based analysis to incorporate intonation and inflection in the sentiment analysis, speaker labelling, etc. The recent features such as topic and language detention and translation have enabled the company to provide business analytics within conversational contexts and expand internationally.
"Audio is among the largest and most data-rich pools of value that companies have left mostly untouched because the tools to analyze and understand that data simply weren't there, weren't fast enough, or were prohibitively expensive to implement," said Madrona Managing Director Karan Mehandru. "
Since its founding, the company has transcribed more than 1 trillion words from over 10,000 years of raw audio data. Among the company's flagship customers, one can see NASA, Spotify, and Citi. Deepgram's speech models narrow the gap between human and AI transcription and are one of the most accurate automatic speech recognition solutions across many languages.
More about Deepgram
Founded in 2015, Deepgram develops AI-based automated speech recognition (ASR) software that uses deep learning-enabled and custom-trained speech models to process voice data. Deepgram's software features include speech-to-text, streaming, and multi-language support.