Automatic Speech Recognition (ASR) biasing helps the agent better understand expected types of user input in each step. It’s especially useful for improving recognition of:

  • Alphanumeric codes
  • Names and name spellings
  • Dates, times, and party sizes
  • Addresses and yes/no responses

You can configure ASR biasing on any flow step to boost transcription accuracy. It will nudge transcription toward expected patterns, like confirmation codes or personal names.

​
Where it appears

In the step editor, you’ll see an ASR biasing panel on the right-hand side when editing a step prompt.

Options include:

  • Alphanumeric – for booking references or confirmation codes.
  • Name spelling – for phonetically spelled names (e.g., “A for apple”).
  • Numeric – for ages or short numbers.
  • Party size – for group bookings.
  • Precise/Relative date – for specific or flexible time requests.
  • Single number – for one-digit responses like menu selection.
  • Time – for spoken times (e.g., “half past eight”).
  • Yes/No – for confirmation-style responses.
  • Address – for location names, postcodes, or street numbers.

Custom keyword fields can also be added to bias transcription toward domain-specific vocabulary.

You can distinguish steps with ASR biasing already applied using the ear icon.

​
When to use it

Enable ASR biasing when:

  • Users are likely to say a structured input like a booking code or date.
  • You’re seeing frequent transcription errors.
  • The agent is making incorrect guesses from unclear audio.

​
Example in a flow

In the flow taken from the example page, the agent is collecting a confirmation code. Because users often say mixed letters and numbers, we toggle Alphanumeric biasing.

This increases the likelihood that phrases like “X9C7G2” are heard and transcribed accurately.

​
Best practices

  • Match ASR bias to the input type, not the prompt phrasing.
  • Combine biasing with few-shot prompting if needed.
  • Always test key steps with real voice input.