2021 Speech Industry Award Winner: Deepgram Applies AutoML to Automatic Speech Recognition

Article Featured Image

Like AppTek, San Francisco-based Deepgram, also joined CallMiner’s Open Voice Transcription Standard (OVTS) as a founding member, but that is just one of the many partnerships it signed in the past year to advance its speech recognition capabilities.

One of those partnerships involved voice specialist Red Box to provide companies with open access to and control of unstructured and structured audio data and the freedom to leverage that data in other applications.

Through the partnership, which was announced in December, Conversa, Red Box’s microservices-based enterprise voice platform, provides real-time audio capture. With deep learning and customizable speech recognition capabilities, Deepgram then transforms this real-time audio stream into transcripts for companies to leverage in artificial intelligence, analytics, and compliance solutions.

That partnership followed a more far-reaching ecosystem expansion that saw Deepgram’s integration with UniMRCP, a move that allows companies to connect speech recognition with interactive voice response (IVR) solutions from companies like Genesys, Cisco Systems, Avaya, and Nuance Communications. UniMRCP is the open-source cross-platform implementation of the Media Resource Control Protocol (MRCP) commonly used for IVRs.

Deepgram’s UniMRCP offering, launched in September, allows companies to capture the reason customers are calling by automatically adjusting to unique audio profiles and by distinguishing between speakers and filtering out background noise.

Around the same time, Deepgram introduced Deepgram AutoML, a tool to streamline artificial intelligence model development to unlock insights from audio data. AutoML, which provides a mechanism by which new AI models can be constructed and tuned automatically, has existed for natural language processing, image, and vision, but Deepgram is the first company to use it specifically for automatic speech recognition.

“Deepgram AutoML reduces the time and effort needed to deploy speech recognition, enabling humans to spend more cycles on overall strategy and processes to successfully integrate AI into their organizations,” Scott Stephenson, CEO and cofounder of Deepgram, wrote in a blog post at the time of the release in August 2020. “Deepgram AutoML pushes the frontier of how AI helps humans evolve next-generation AI.”

And just this past June, Deepgram introduced conversational AI, sales and support enablement, and real-time streaming to enable the creation of virtual agents for specific high-volume voice tasks, like billing, support, add-on sales, compliance, or identity verifications.

Deepgram’s conversational AI feature reduces the response lag to less than 300 milliseconds (among the best in the industry), improves word error rate by up to 50 percent, and reduces noise and filters out crosstalk to help focus on the key terms and phrases for intent and understanding.

“The promise of AI has consistently been to solve problems and deliver experiences in ways that are faster, cheaper, and better. At Deepgram, AI is in our DNA. We’re an automatic speech recognition company focused on solving the inherent challenges of speech technology and powering the next generation of voice-enabled applications,” Stephenson explained.

On the heels of a $25 million Series B funding round earlier this year, Deepgram also expanded its base language models to include not just English but also French, Hindi, Korean, Portuguese, Russian, Spanish, and Turkish; improved its platform’s hardware performance by three times; and added a data center dedicated to research. The company is also poised for impressive year-over-year revenue and employee growth, something not many companies can say amid COVID-19-induced hardships.

“With the ASR industry rapidly expanding, more and more companies are recognizing speech as a powerful tool that unlocks invaluable insights and helps shape state-of-the-art voice experiences,” said Mehul Patel, vice president of products at Deepgram. “At Deepgram, we are constantly working to ensure that we’re providing customers with the highest-quality speech recognition platform on the market. We’re poised for strong, continued growth as a result of our recent momentum and look forward to expanding our offerings and customer base in 2021 and beyond.”

SpeechTek Covers
for qualified subscribers
Subscribe Now Current Issue Past Issues