> For the complete documentation index, see [llms.txt](https://learning.contextqa.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/metrics-reference.md).

# Voice AI Metrics Reference

Understand ContextQA audio judgments, measured latency, interruptions, dead air, talk ratio, and business criteria for voice agent evaluation.

> **Documentation owner:** [Deep Barot](https://contextqa.com/academy/) · **Last updated:** November 11, 2026

## Quick answer

ContextQA distinguishes LLM-judged voice criteria from deterministic audio measurements. Use judgment for contextual qualities such as tone or language, and measured criteria for timing and interaction signals. Enable the needed audio metrics in the project and review their definitions and evidence before using them as release gates.

## Key takeaways

* An available audio criterion may remain off until you enable it.
* Measured signals and LLM judgments answer different questions.
* Calibrate informational measures before using them to block releases.

## Standard voice criteria

The inspected project included these LLM audio criteria: **Audio Quality**, **Language Match**, **Tone Appropriate**, **Response Audibility**, **Speech Rate**, **Turn Taking**, **VAD Responsiveness**, and **Dead Air**.

Use their descriptions to understand the judgment. For example, an LLM assessment of dead air is a contextual judgment; it is not interchangeable with a deterministic duration measurement. Likewise, a Speech Rate judgment and a measured Speaking Rate Band can provide different evidence.

## Additional measured audio criteria

The live **Available audio criteria** panel states that criteria are off until enabled and that enabled criteria score every voice case in the project. Check this scope before enabling a criterion that only makes sense for a subset of your workflows.

| Criterion                 | What to inspect in a review                                      |
| ------------------------- | ---------------------------------------------------------------- |
| Measured Response Latency | The measured response interval and its definition                |
| Agent Interruptions       | Instances where the agent overlaps or interrupts the caller      |
| Yields When Interrupted   | Whether the agent stops appropriately when the caller intervenes |
| Measured Dead Air         | Silence intervals against the project’s acceptance policy        |
| Talk Ratio                | The balance of speaking time for the intended interaction        |
| Speaking Rate Band        | Whether measured speech speed fits the configured band           |
| Line Quality              | The evidence for connection or signal quality                    |
| Echo Free                 | Echo-related evidence in the audio                               |
| Speech Quality Score      | Informational output until calibrated for the use case           |

![Available measured voice criteria with controls for enabling latency, interruptions, dead air, and audio quality checks](https://4255556216-files.gitbook.io/~/files/v0/b/gitbook-x-prod.appspot.com/o/spaces%2FKlyaXxb76z3Dn47TrZkU%2Fuploads%2Fgit-blob-be0dae1c5d90df2446769810fed4b156b720a2d7%2Fmeasured-voice-metrics.jpg?alt=media)

These descriptions guide review; they do not specify undocumented formulas, units, defaults, or threshold values. Open the criterion’s current configuration for its exact definition. In particular, establish what timestamps bound a latency measurement before comparing it with another tool’s latency figure.

## Additional audio judgments

The panel also offered **Prosody and Expressiveness**, **Intelligibility**, and **Entity Pronunciation** as LLM audio criteria. Use them when the requirement depends on naturalness, understandable speech, or pronunciation of important entities.

Entity Pronunciation asks about how an entity is spoken. It does not establish that the entity’s value is correct. A clearly pronounced wrong date still fails the date-accuracy requirement.

## Build a balanced metric set

For a booking workflow, combine relevant audio criteria with Multi-Turn Context, a custom final-details rubric, and backend verification. For a support agent, use criteria appropriate to correct resolution, grounded answers, and escalation. Avoid enabling unrelated measures merely to increase the number of scores.

Review a small set of known good and known bad calls manually. Confirm that the metric distinguishes the cases you care about. Document tolerances and exceptions before expanding the suite. Keep Speech Quality Score informational until its behavior has been calibrated against representative calls.

Continue with [dead air diagnosis](https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/dead-air-and-latency.md), [interruptions and context](https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/interruptions-and-context.md), and [custom metric configuration](https://learning.contextqa.com/ai-agent-testing/agent-evaluation/metrics-and-judges.md).


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation by asking a question.

Perform an HTTP GET request on the following URL with the `ask` and `goal` query parameters:

```
GET https://learning.contextqa.com/ai-agent-testing/voice-ai-testing/metrics-reference.md?ask=<question>&goal=<user_goal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is what the user is ultimately trying to achieve, the reason they need the answer. Sharing it helps GitBook give you a better, more relevant answer. A goal is most helpful when it describes the outcome the user wants rather than restating the question. For example, with `ask=how do I create an API token`, a goal like `automate deployments from our CI pipeline` lets GitBook tailor the answer to that use case.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.
