February 2021 Summaries

Black History Month is a time for reflection and celebration of African American culture, including the dialect known as African American Vernacular English (AAVE). AAVE has evolved over centuries from influences of African languages and creoles, and it continues to be influenced by Standard American English (SAE) and other regional and community-based dialects. Despite its complexity and distinctiveness, AAVE is often dismissed or misunderstood. One challenge for speech recognition technology is code-switching, where speakers switch between dialects or languages in conversation. Currently, most speech recognition systems focus on SAE, leading to higher error rates when transcribing other dialects like AAVE. It's crucial for companies in the speech recognition space to improve accuracy for diverse dialects and embrace the full spectrum of human speech.

Nuance, a well-established brand in speech recognition, has recently acquired Saykara to expand its medical transcription business. While it is good at non-real time transcriptions like medical records, its core architecture based on the Hidden Markov Model sacrifices speed for accuracy. Deepgram, on the other hand, uses a different architecture that enables real-time AI and does not compromise between accuracy, speed, cost, or scalability. For real-time streaming, Deepgram's transcription lag is under 300 milliseconds.

Deepgram has raised $25 million from Tiger Global and others to advance speech recognition technology in response to the evolving market landscape. The COVID-19 pandemic accelerated digital transformation initiatives, highlighting the need for voice enabled applications as a necessity rather than a luxury. Deepgram aims to set the standard for building such applications and help enterprises differentiate themselves through tailored approaches to speech.