Published Mar 23, 2020

Speech recognition to say it just right

Catherine Breslin delves into the vast applications and design challenges of speech recognition technology, from enhancing accessibility and transcription to refining virtual assistants' conversational abilities and tackling noise and accent variability for improved system performance.
Episode Highlights
Practical AI logo

Popular Clips

Episode Highlights

  • Design Choices

    Designing virtual assistants involves making critical choices about how they interact with users. explains that these systems must balance between providing random responses and seeking specific user input, such as when playing music or delivering weather forecasts 1. The goal is to minimize user frustration by keeping conversations concise, while accurately interpreting user intents like playing music or turning on lights. Breslin notes, "The assistants are not that great at conversation yet," emphasizing the need for simplicity in design 1.

       

    Conversation Dynamics

    Managing conversations in virtual assistants is complex due to the limitations of current language understanding technologies. Breslin discusses how traditional systems modeled conversations as flowcharts, which can be rigid and limiting 2. Advances now allow for more flexibility, enabling users to express their needs in a single utterance rather than multiple turns. However, tracking conversation states over time remains challenging, particularly when users refer back to previous topics. "There's a lot of complicated language-related technology to do that tracking over time," Breslin remarks, highlighting ongoing challenges 2.

Related Episodes