nemo_voice_agent.utils.voice_prompts

Shared voice-realization prompt fragments.

These constants tell an LLM how to speak — concise plain-text output, characters-at-a-time spelling for identifier codes, no canonical numeric forms alongside spelled-digit forms. They apply to any voice agent regardless of whether it’s running under the evaluation harness or the production server, and to both the agent side and the simulated-user side.

This module is import-cheap (no pipecat/NeMo/torch dependencies) so it can be pulled into any layer of the codebase without dragging in heavy imports.

For agent-behavior guidelines tied to specific eval-harness tools (e.g. when to call EndConversationTool), see nemo_voice_agent.evaluation.scenarios.END_CONVERSATION_GUIDELINE. Those are evaluation-specific and live alongside the scenario base classes.

Module Contents

Data

GENERAL_PROMPT

VOICE_ALPHANUMERIC_RULE

API

nemo_voice_agent.utils.voice_prompts.GENERAL_PROMPT = "Keep your responses concise and conversational since they will be spoken aloud....

nemo_voice_agent.utils.voice_prompts.VOICE_ALPHANUMERIC_RULE = 'When speaking confirmation numbers, flight numbers, airport codes, user IDs, ph...