Global ASR

Improve speech recognition accuracy with keyphrase boosting and transcript corrections.

Level 2 — Lesson 6 of 8 — Configure speech recognition to reduce misrecognition and improve routing.

ASR determines what the agent hears. Every downstream system depends on it: retrieval, routing, response control, handoff logic. Small transcription errors here silently break otherwise well-designed agents. Global ASR is where you correct systematic listening problems at the source.

What global ASR is for

Reduce clarifications

Minimize unnecessary clarification questions

Improve routing

Increase routing accuracy for calls

Stabilize retrieval

Improve Knowledge topic selection for spoken queries

Fix transcription

Correct recurring transcription errors at the source

It is not a substitute for good KB design or clear agent behavior rules. It improves transcription accuracy — it does not add new knowledge to the agent.

Which tool to use

Two essential tools

Keyphrase Boosting

Keyphrase Boosting biases the ASR model toward recognising specific words or phrases more reliably.

When to use it

Use cases

• A word is important for routing or actions
• Misrecognition causes the agent to take the wrong path
• Users consistently repeat themselves for the same term

Typical candidates

Department names:

• “front desk”
• “billing”
• “room reservations”

Product or service names:

• “bell desk”
• “late checkout”

Action-critical verbs:

• “cancel”
• “confirm”

How to configure

1. Add the keyphrase

   Enter it exactly as you want it recognized

2. Set bias strength

   Start with Default bias strength

3. Adjust if needed

   Increase toward Maximum only if errors persist

Important guidance:

• Bias strength is cumulative. Over-biasing many terms can degrade overall accuracy
• Avoid boosting generic words like “room” unless they are routing-critical
• Prefer multi-word phrases over single words when possible
• Example: If callers asking for the “bell desk” are frequently routed incorrectly, boost “bell desk” rather than boosting “desk”

Transcript Corrections

Transcript Corrections rewrite text after transcription, mapping common errors to the intended phrase.

When to use it

Use cases

• ASR consistently hears the same wrong thing
• Accents or homophones cause predictable errors
• Brand or proper names are repeatedly mangled

Typical examples

Homophones:

• “copper” → “hopper”

Accent-driven substitutions:

• “reef fund” → “refund”

Brand names:

• “Hopper Consumer” → “Hopper”

How to configure

1. Choose match type

   Select substring or regex

2. Define misheard expression

   Enter what ASR incorrectly transcribes

3. Define correction

   Enter the intended replacement text

Important guidance:

• Prefer narrow matches over broad regex
• Avoid corrections that could unintentionally rewrite unrelated phrases
• If unsure, test in Sandbox with real call audio before promoting
• Example: If “cancel” is sometimes transcribed as “council,” add a correction rather than increasing bias on “cancel” alone

Check your understanding

A practical workflow

1. Review recent calls

   Look for calls where:

   • The agent asked for repetition
   • The wrong KB topic was selected
   • Routing or handoff failed unexpectedly
2. Identify the problem

   Find the specific word or phrase that failed

3. Choose your tool

   Decide:

   • Boost if the word is correct but under-recognized
   • Correct if the word is consistently misheard as something else
4. Apply minimal change

   Make the smallest possible adjustment

5. Test in Call

   Verify the improvement

6. Promote carefully

   Only promote after confirming improvement

Common mistakes

Avoid these pitfalls:

• Boosting everything “just in case”
• Using Transcript Corrections to fix KB design problems
• Applying maximum bias globally without testing
• Forgetting that local (flow-level) ASR settings override global ones

Check your understanding

Verification checklist

Verification is complete when:

• Previously misheard phrases are transcribed correctly
• The agent asks fewer clarification questions for known terms
• Routing and KB selection stabilise in live call testing
• Behavior remains consistent after promotion between environments

Global ASR tuning is iterative. Small, deliberate adjustments here pay off more than large rewrites elsewhere.

Try it yourself

1. Challenge: Diagnose and configure an ASR fix

   A caller says “I’d like a refund” but ASR consistently transcribes it as “I’d like a reef fund”. The refund_request topic therefore never triggers.

   Answer:

   1. Should you use Keyphrase Boosting or Transcript Corrections — and why?
   2. Write the full configuration.

   Hint

   Ask yourself: is ASR hearing the right word but recognising it poorly, or is it consistently mishearing one thing as another?

   Example solution

   1. Transcript Corrections — because ASR is consistently mishearing “refund” as “reef fund”. This is a predictable substitution error, not an under-recognition problem.

   2. Configuration:

      • Match type: Substring
      • Misheard expression: reef fund
      • Correction: refund

Check your understanding

← Previous: Audio management

Lesson 5 of 8

Next: Variants →

Lesson 7 of 8