> For the complete documentation index, see [llms.txt](https://docs.openagrinet.global/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.openagrinet.global/technical/ai-layer/guidelines-gemma-testing/voice-llm-guidelines.md).

# Voice LLM Guidelines

#### Human Evaluation Guidelines (Voice only)

* Voice Channel Specifics (Voice)
* brevity (1-4)
* voice\_ready (0-1)
* tone\_naturalness (1-4)
* speech\_accuracy

brevity (1-4): Voice-appropriate response length.

* Metric Description: The answer should be short and spoken in under 30 seconds. If a comprehensive answers is long, then it just says the crux/first part of the answer and the rest of the answer is asked back as followup
* Reasoning:\
  Expected behaviour is that the response is short and is spoken under 30 seconds.  Score 4 if the response is compact and can be spoken in under 30 seconds and has a followup for the rest of the answer Score 3 if the bot gives a short answer but there is no proper followup question for the rest of the answer. Score 2 if the answer is around 30 seconds to 60 seconds. Score 1 if the answer is more than 60 seconds. <br>

voice\_ready\_text (0-1): Absence of text artifacts in spoken output.

* Metric Description: Output should be all speech and devoid of any text/chat artifacts like ‘:’, ‘\*’ that is spoken out by the TTS system.&#x20;
* Reasoning: Score 1 if the text is clean of markdown and doesn’t include any ‘non-spoken’ text. Score 0 if the AI literally says artifacts like "Asterisk," "Bullet point," or "Bold."\
  \
  Also point out the ‘word’/’symbol’ that were spoken

tone\_naturalness (1-4): Respectful, warm, and conversational style.

* Metric Description: Tone should be warm and friendly while sounding natural &#x20;
* Reasoning: Score 4 if the voice sounds natural and voice tone is friendly and helpful Score 1 if the tone is cold, dismissive, or overly robotic. Score 2, 3 depending on how robotic and unnatural it sounds &#x20;

**Speech\_accuracy**&#x20;

* Metric Description: Number of words spoken incorrectly or pronounced incorrectly. This is just a number of words that were spoken wrong/pronounced wrong.&#x20;


---

# 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://docs.openagrinet.global/technical/ai-layer/guidelines-gemma-testing/voice-llm-guidelines.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.
