Standard Metrics

Every call automatically gets analyzed for 9 standard metrics. Audio-based metrics require a recording_url; transcript-based metrics only need a transcript.

Metric reference

latency_ms — Response Latency

Time between the user stopping speaking and the AI starting to respond, per turn.

Requires: Audio recording

1{
2 "metric_key": "latency_ms",
3 "status": "computed",
4 "result_data": {
5 "average_ms": 312,
6 "p95_ms": 540,
7 "turns": [
8 { "turn": 1, "latency_ms": 295 },
9 { "turn": 2, "latency_ms": 328 }
10 ]
11 }
12}

user_interruptions — User Interruptions

Times the user spoke over the AI.

Requires: Audio recording

1{
2 "metric_key": "user_interruptions",
3 "status": "computed",
4 "result_data": {
5 "count": 2,
6 "interruptions": [
7 { "at_second": 12.4, "duration_ms": 600 },
8 { "at_second": 34.1, "duration_ms": 1200 }
9 ]
10 }
11}

ai_interruptions — AI Interruptions

Times the AI spoke over the user.

Requires: Audio recording

1{
2 "metric_key": "ai_interruptions",
3 "status": "computed",
4 "result_data": {
5 "count": 1,
6 "interruptions": [
7 { "at_second": 22.8, "duration_ms": 400 }
8 ]
9 }
10}

silence_detection — Silence Detection

Silent periods per channel (AI and user separately).

Requires: Audio recording

1{
2 "metric_key": "silence_detection",
3 "status": "computed",
4 "result_data": {
5 "ai_channel": { "count": 3, "total_ms": 4200 },
6 "user_channel": { "count": 1, "total_ms": 900 }
7 }
8}

stop_time_after_interruption — Stop Time After Interruption

How quickly the AI stops talking after the user interrupts.

Requires: Audio recording

1{
2 "metric_key": "stop_time_after_interruption",
3 "status": "computed",
4 "result_data": {
5 "average_ms": 180,
6 "events": [
7 { "at_second": 22.8, "stop_time_ms": 180 }
8 ]
9 }
10}

transcription_accuracy — Transcription Accuracy

Word Error Rate (WER) comparing the provided transcript against the generated one.

Requires: Both a provided transcript and audio (for generated transcript comparison)

1{
2 "metric_key": "transcription_accuracy",
3 "status": "computed",
4 "result_data": {
5 "wer": 0.04,
6 "cer": 0.02,
7 "reference_words": 214,
8 "hypothesis_words": 218
9 }
10}

voice_change_detection — Voice Change Detection

Detects significant changes in the AI agent’s voice characteristics during the call.

Requires: Audio recording

1{
2 "metric_key": "voice_change_detection",
3 "status": "computed",
4 "result_data": {
5 "changes_detected": false,
6 "events": []
7 }
8}

voice_tone_clarity — Voice Tone & Clarity

Scores the clarity and naturalness of the AI’s voice on a 1–5 scale.

Requires: Audio recording

1{
2 "metric_key": "voice_tone_clarity",
3 "status": "computed",
4 "result_data": {
5 "score": 4,
6 "label": "Good"
7 }
8}

word_per_minute — Word Per Minute (WPM)

AI agent’s speech speed.

Requires: Audio recording

1{
2 "metric_key": "word_per_minute",
3 "status": "computed",
4 "result_data": {
5 "wpm": 148
6 }
7}

Metric statuses

StatusMeaning
pendingAnalysis queued or in progress
computedResult available in result_data
failedAnalysis encountered an error
no_recordingMetric requires audio but none was provided