2022 Speech Industry Award Winner: Deepgram Is the Startup Speech-Enabler

Some could argue that Deepgram reinvented automatic speech recognition (ASR) with a complete, deep learning model that yields faster, more accurate transcriptions with lower hardware and usage costs. Now the vendor is giving away $10 million in free speech recognition to developers.

The startup generously created the Startup Program to provide select companies and individual developers with custom-trained speech models to help them harness the power of speech quickly, easily, and economically. It is bestowing the $10 million in free credits, with a specific focus on startups in education and employee and/or customer experience.

The Deepgram Startup Program offers a range of perks, including up to $100,000 in Deepgram ASR to use over 12 months; exclusive access to training sessions, materials, new features, and speech models; opportunities for developers to showcase their solutions and companies; dedicated customer success managers; and joint go-to-market initiatives.

“We believe in a world where every voice is heard and understood,” Deepgram CEO Scott Stephenson, who cofounded the company in 2015, said in a statement. “Developers playing and prototyping with our technology is one of the fastest paths to making voice a de facto mode for communications. We’re just scratching the surface of how voice can transform businesses and can’t wait to see how these awesome practitioners leverage automatic speech recognition in the real world.”

The program comes as COVID-19 accelerated companies’ interest in and need for speech technologies to stay connected to employees and customers, according to Stephenson. It follows a funding round in early 2021 that raised $25 million.

To coincide with the program, Deepgram added platform features, including a developer console and software development kits for Python and Node.js. The new Developer Console is a graphical user interface that enables developers to access the full range of Deepgram ASR services. One of the console’s key features, Missions, provides a learning path for getting started with Deepgram, with tasks to accomplish and rewards along the way. The SDKs enable developers to start developing applications with Deepgram’s speech-to-text platform to transcribe both real-time streaming and prerecorded audio, as well as experiment with new applications. Deepgram’s new SDKs include libraries, documentation, code samples, processes, and guides.

But industry newcomers aren’t the only ones benefiting from Deepgram’s recent innovations. The company this year also expanded to 23 the number of languages and dialects that its speech recognition supports. The current lineup of supported languages and dialects includes Dutch; U.S., U.K., Australian, and Indian English; European and Canadian French; German; Hindi; Indonesian; Italian; Japanese; Korean; simplified and traditional Mandarin Chinese; European and Brazilian Portuguese; Russian; European and Latin American Spanish; Swedish; Turkish; and Ukrainian.

Mehul Patel, vice president of products at Deepgram, says the company developed the new language models using transfer learning backed by its End-to-End Deep Learning architecture and training with real-world audio datasets.

Deepgram also expanded its transcription for three new use cases: earnings calls, voicemail, and video. These models have been trained on real conversations, enabling them to better transcribe jargon, accents, dialects, code-switching, terminology, and alphanumerics.