Our Chief Data Scientist Vijay Gurbani Ph.D and his team debuted this white paper last month at the VOICE & AI 2023 conference and now we’re making it available to you!

Even with the best automated speech recognition (ASR) solution, there is a likelihood that the performance of the system for a particular use case or specialized domain does not meet requirements. While human-level speech recognition accuracy is estimated at > 95%, many applications require accuracies closer to 99%.

Moreover, domain expertise is routinely assumed. For instance, an expert in transcribing legal documents lacks the domain expertise to accurately transcribe medical documents and vice versa. The same disadvantage applies to voice applications as well. Bespoke speech recognition models have a powerful advantage in the area of transcription accuracy for domain-specific use cases. Cost is also an important factor to consider when evaluating options since highly-accurate general purpose ASR models can be expensive to license and manage for large deployments generating large data sets.

Vijay has authored or co-authored over 70 papers in peer reviewed journals, 5 books, 19 Internet Engineering Task Force (IETF) RFCs, and been granted 8 patents by the U.S. Patent Office.

Download the white paper now to get the in-depth story about custom speech recognition models.