Response Schemas

Timeline Entry

Each element under timeline contains multimodal signals for a single utterance.

string

Unique identifier for the utterance

string

Speaker identifier from diarization

number

Start time in seconds

number

End time in seconds

string

Transcribed text from speech recognition

number

ASR confidence score (0-1)

object

Acoustic metrics including:

Pitch/energy statistics
Pauses
Spectral features
Commitment/passion markers
Lexical markers
Hesitation flags

string

Natural-language descriptor (e.g., “COMMITTED, PASSIONATE, emphatic”)

object

Full emotion analysis including:

Emotion scores (joy, sadness, etc.)
PAD scores (valence, arousal, dominance)
Dominant emotion
Confidence

object

Structured sentiment analysis

number

Sentiment polarity (-1 to 1)

number

Subjectivity score (0-1)

object

Cognitive analysis including:

Engagement score
Cognitive load level
Cognitive signals

number

Derived commitment score from prosody

string

Commitment level classification

number

Derived passion score from prosody

string

Passion level classification

number

Conviction score derived from prosody and acoustic reasoning

object

Lexical analysis including:

Hedges
Fillers
Questions
Disfluency rate
Self-repair flags

string

Narrative produced explaining how acoustic cues map to cognition/emotion

number

Average confidence across all agents (0-1)

ConversationSummary Schema (Reference)

Fields:

Field	Type	Notes
`overview`	string	Executive-level synopsis.
`key_insights`	string[]	Bullet-ready highlights (timestamp references encouraged).
`notable_patterns`	string[]	Trends across the conversation.
`recommendations`	string[]	Actionable next steps.
`what_went_well`	string[] \| null	Populated when the user explicitly asks “what went well.”
`what_went_wrong`	string[] \| null	Populated on “what went wrong” requests.
`speaker_dynamics`	`[{speaker, description}]` \| null	Characterization per speaker.
`conversation_quality`	`“excellent"	"good"	"fair"	"poor”`	Defaults to `"good"`.

When contexts such as “call out what went well” are provided, those optional arrays will contain 3-5 evidence-backed bullets.

Insights

Aggregated insights from the fusion service.

object

Sentiment statistics:

mean: Average sentiment score
std: Standard deviation
trend: Sentiment trend (increasing/decreasing/stable)

array

Top three dominant emotions (strings)

string

Engagement pattern: increasing, decreasing, or stable

array

Curated subset of detected moments (MomentEvent objects)

object

Profiles for each speaker

object

Analysis of conversation dynamics

object

Overall cognitive profile

Transition Event

Affective state transition detection.

string

Unique transition identifier

integer

Timestamp in milliseconds

object

Previous affective state

object

New affective state

array

Array of utterance IDs that contributed to this transition

object

Valence/arousal/engagement deltas that triggered the transition

number

Confidence score (0-1)

string

Stability indicator

Moment Event

Key moment detection.

string

Unique moment identifier

string

Moment category: objection, cta_offered, cta_accepted, cta_rejected, topic_shift

string

Speaker identifier

integer

Start timestamp in milliseconds

integer

End timestamp in milliseconds

array

Array of associated utterance IDs

string

Summary of the moment

array

Additional labels

number

Confidence score (0-1)

Conversation Segmentation

Segmentation of conversation into phases.

array

Array of ConversationPhase objects

integer

Total number of phases

string

Explanation of segmentation approach

string

Narrative arc description

array

Array of key transitions (TransitionEvent objects)

Conversation Phase

Individual phase within a segmented conversation.

string

Unique phase identifier

string

Name of the phase

number

Start timestamp in seconds

number

End timestamp in seconds

string

First utterance ID in this phase

string

Last utterance ID in this phase

integer

Number of utterances in this phase

string

Theme of the phase

string

Summary of the phase

string

Emotional tone of the phase

array

Key moments within this phase (MomentEvent objects)

object

Transition signal that led to this phase (TransitionEvent)

Error Response

Standard error response format.

string

Error message describing what went wrong

Example:

{
  "detail": "Invalid or missing API key"
}

Live Session Response

Response from starting a live session.

string

Unique session identifier

string

ISO 8601 timestamp when session expires

string

Current session status

integer

Next expected chunk sequence number (zero-based)

Metrics Response

Observability metrics for dashboards.

object

Event counts by type:

status: Number of status events
final_transcript: Number of final transcript events
emotion: Number of emotion events
Additional event types as key-value pairs

object

Average agent latency per utterance type (milliseconds)

Example:

{
  "events": {
    "status": 45,
    "final_transcript": 12,
    "emotion": 12
  },
  "agent_latency_ms": {
    "utterance": 215.34
  }
}

Introduction

Endpoints

Cookbooks

Reference

Timeline Entry

ConversationSummary Schema (Reference)

Insights

Transition Event

Moment Event

Conversation Segmentation

Conversation Phase

Error Response

Live Session Response

Metrics Response

​Timeline Entry

​ConversationSummary Schema (Reference)

​Insights

​Transition Event

​Moment Event

​Conversation Segmentation

​Conversation Phase

​Error Response

​Live Session Response

​Metrics Response

Timeline Entry

ConversationSummary Schema (Reference)

Insights

Transition Event

Moment Event

Conversation Segmentation

Conversation Phase

Error Response

Live Session Response

Metrics Response