Production-grade acoustic vocal emotion and frustration detector for voice AI agents. Calculates emotional valence and acoustic arousal to trigger de-escalation protocols.
Single tool with critical definition gaps. Tool name 'detect_audio_emotion' lacks action verb prefix (should be 'analyze_audio_emotion' or 'detect_emotion_from_audio'). Description is adequate (89 chars, within 10-1024 range) but parameters lack descriptions entirely, none of the three input parameters (pitch_hz_variance, speech_rate_syllables_sec, amplitude_jitter) have descriptions explaining what they control or their valid ranges. Input schema present with types and defaults, but no parameter-level documentation. Output schema not documented, LLM cannot predict response structure. No error handling guidance. Tool definition visible in mcp_server.py but parameter descriptions completely absent.
Detects customer emotional valence and vocal frustration from speech prosody.
All input parameters lack descriptions. pitch_hz_variance, speech_rate_syllables_sec, and amplitude_jitter have no explanation of what they control, valid ranges, or units beyond the name itself.
Output schema not documented. Response includes 8 fields (detection_id, acoustic_arousal_score, emotional_valence_score, pitch_variance_hz, speech_rate_syllables_sec, frustration_detected, recommended_agent_persona_adaptation, barge_in_priority_boost) but LLM has no schema to predict structure or field types.
Tool name lacks action verb prefix. 'detect_audio_emotion' is passive; should start with clear verb like 'analyze_audio_emotion' or 'detect_emotion_from_audio' to signal intent to LLM.
No error handling or recovery guidance. Tool returns hardcoded logic with no validation of input ranges. If pitch_hz_variance is negative or speech_rate is 0, behavior is undefined.
| Scored | Grade | Overall | Spec posture | Rubric |
|---|---|---|---|---|
| 2026-09-23 | F | 41 | 2026-07-28+ | v2 |
Parameter defaults (85.4, 6.2, 0.045) are unexplained. LLM cannot determine if these are sensible defaults or placeholder values. No guidance on typical ranges or when to override.