Unlocking the Largest Data Source in the Enterprise: VoiceZach DeWitt | March 18, 2020
Locked Treasure Trove of Voice Data
Every year, Apple’s customer service team speaks with millions of customers to help troubleshoot devices. Uber, Lyft, and Doordash conduct millions of phone interviews to screen for new drivers. JPMorgan Chase speaks to millions of households and small businesses to help with financial planning. Collectively, enterprises converse for trillions of minutes annually. There is a treasure trove of information-rich data waiting to be unlocked from all of these conversations.
Voice is one of the biggest trends in enterprise technology. However, today’s speech recognition technology has been optimized for one-way, short-form conversations like those delivered to a virtual assistant. Unlike Siri and Alexa commands, enterprise communication is messy. Most employees aren’t speaking clearly and slowly into a high-fidelity microphone. Instead, meetings and calls usually have multiple people with different accents speaking over each other far away from the microphone.
Making sense of verbal communication for enterprises is a massive opportunity, but it’s nascent given technical challenges and prohibitive costs. Enter Deepgram.
Turning Voice into Value through Deep Learning
Advancements in natural language processing (NLP) are helping enterprises understand intention and perception from emails and other forms of written communication. However, the holy grail has been to understand verbal communication in a similar way. Deepgram has taken an entirely new approach to automatic speech recognition (ASR) by coupling robust deep learning models with specialized data infrastructure.
Through Deepgram’s API, data scientists and developers can convert messy, unstructured audio data into accurate and structured transcriptions in real-time. Deepgram builds custom end-to-end deep learning models for every customer. This architecture enables Deepgram to achieve ~95% accuracy vs. ~72% accuracy from incumbent competitors. In addition to significantly improved accuracy, Deepgram’s technology enables better pricing, higher reliability and superior performance at scale.
Inspired by Dark Matter
Deepgram’s founders are uniquely positioned to build high-performance voice infrastructure. The concept came while both founders were Ph.D. students researching the creation of dark matter two miles underground. It was after that research experiment they discovered there wasn’t a tool available that would help process recordings and pinpoint valuable timestamps. When we met Scott Stephenson and the team, we were blown away by their customer obsession, technical prowess, and commercial ambition.
Deepgram publically launched its speech engine two years ago and its customers depend on its technology for mission-critical analysis and workflow. To date, Deepgram has processed over one billion minutes of voice data. When we spoke with Deepgram’s customers, we were amazed at how thorough they had been in testing Deepgram vs. competitors. Deepgram consistently exceeded the competition across multiple performance dimensions.
Deepgram will enable a new generation of enterprise applications and services. We are just scratching the surface of what information and signals can be extracted from voice data. Deepgram is built for developers and is architected to power the full spectrum of companies ranging from a small startup to a multinational Fortune 100 company.
Wing’s Partnership with Deepgram
We are thrilled to announce that Wing has led Deepgram’s $12M Series A financing. We are proud to invest alongside our friends at NVIDIA, SAP.iO, Compound Ventures, and Y Combinator. I have joined Deepgram’s board and am excited to work with the team to help enterprises better understand and leverage their voice data.
If you would like to learn more about Deepgram, please visit the company website. Deepgram is hiring and if you are interested in working on hard problems with an incredible team, please visit Deepgram’s Careers page.
Share this Article