​
Overview

PolyAI uses advanced Automatic Speech Recognition (ASR) models to convert spoken language into written text. To improve the recognition of critical domain-specific terms, use the Keyphrase Boosting feature.

​
Key benefits

  • Improved SLU accuracy: Increases the likelihood of accurately recognizing critical terms.
  • Domain adaptability: Ensures the agent recognizes industry-specific terminology.
  • Enhanced user experience: Reduces misunderstandings and errors, leading to more effective interactions.
  • Easy management: Manage and update keyphrases in the UI.

​
How it works

Keyphrase Boosting biases the ASR model toward recognizing specific words and phrases. By curating a list of keyphrases relevant to your domain, you can improve transcription accuracy for those terms, resulting in better inputs for the Language Learning Model (LLM) and improved agent performance.

​
Getting started

​
Configuring Keyphrase Boosting

  1. Access the Speech Recognition page: Navigate to the Speech Recognition section.
  2. Manage keyphrases:
    • In the Keyphrase Boosting tab, add, edit, or remove keyphrases.
    • Use the Keyphrase column to input domain-specific terms.
  3. Set bias strength:
    • Adjust the bias strength for each keyphrase using the slider.
    • Options range from Default (moderate priority) to Maximum (high priority).
  4. Save changes: Once configured, the updated keyphrases will be applied immediately to improve ASR recognition.

​
Bias strength configuration

  • Default: Balances recognition accuracy with overall ASR performance.
  • Maximum: Prioritizes keyphrases for improved accuracy but may impact general performance.
Important: When both global and local biasing are applied, local settings take precedence.

​
Diacritics

If your agent is operating in a language that uses diacritics — such as č, ć, š, ž, đ — we recommend reaching out to platform-support@poly-ai.com for setup guidance. These characters are common in languages like Croatian, Serbian, Bosnian, Slovak, Czech, Slovenian, and others. Some ASR models can misinterpret these letters, reducing accuracy for names and other proper nouns.