Best Practices for Building a Custom-Made ASR: What You Should Know

Presented by: 

Chatbots powered by LLMs (Large Language Models) like ChatGPT are more powerful than ever. However, their application in the customer service domain introduces valid concerns regarding privacy and dependability. In our experience, domain-specific Automatic Speech Recognition (ASR) engines combined with Natural Language Understanding (NLU) models provide a robust solution for enhancing customer interactions.

A highly performant ASR system is key to unlocking downstream natural language processing capabilities in many enterprise scenarios. Domain-specific ASR engines accommodate specialized jargon, evolving domain requirements and specific user needs. Further, they can adapt to challenging channel-specific acoustic environments. Hence, custom-made or "bespoke" models reach a higher accuracy level compared even to state-of-the-art domain-general models for specific contexts. Not only do bespoke models require less data and hardware for training and deployment, but they are also economical. On a larger scale, precise domain-general ASR models come with substantial licensing and utilization costs.

Join us to get valuable insights into the best practices of training Bespoke ASR models to provide reliable in-house Natural Language Processing (NLP) solutions. This webinar will delve into the inner workings of ASR engines and discuss domain-specific training requirements and considerations. Additionally, we’ll explore testing, monitoring and evaluation metrics for domain-specific models.

Presented by Nikhita Sharma, Data Scientist at Vail Systems

Nikhita Sharma is a Data Scientist at Vail Systems, Chicago. She has experience in applied machine learning research, primarily focused on conversational AI, Automatic Speech Recognition (ASR), and Natural Language Processing (NLP). Nikhita has a background in software engineering and enjoys working on a variety of data science projects. She has a master’s in computer science from Iowa State University.

 

Our Moderator for the Webinar will be Pete Erickson, Founder & CEO of Modev. Pete is a musician and entrepreneur who believes that human connection is vital in the era of digital transformation. He founded Modev in 2009 to turn this belief into a mission of bringing the tech industry together. Today, Modev organizes some of the world’s top events, such as VOICE & AI.  Modev is a global team of more than 40 creatives, marketers and producers. Pete has been a contributor nationally at NPR, CBS, NBC and serves as the on-air “Tech Expert” for Fox 5 DC where he appears regularly to provide commentary on technology news that impacts consumers.