Skip to content

Contexta Analyzer

tim-contexta edited this page Aug 12, 2020 · 7 revisions

Contexta Analyzer analyzes transcriptions of (business) phone calls and adds additional (meta) fields to the transcript that give more insights about the conversation. Meta data fields can be performance benchmarks for agents or give more insights about the content of the conversation.

Access to Analyzer

Clients with a Contexta Analyzer subscription can access the transcription and scores via the /transcripts/:transcriptId endpoint.

Analyzer fields

The meta data can be placed on three different levels in the transcription:

  1. Meta data about the full conversation
  2. Meta data for every speaker
  3. Meta data for every segment

Full conversation metrics

Fields containing information about the full conversation.

metadata.keywords

This field contains the most important keywords or phrases from the conversation, using a Contexta NLP model.

  • type: array

example: { [ "compare energy online", "last year", "variable costs", "get a new energy contract", "what's the price" ] }

metadata.total_duration

Total duration of the conversation in seconds.

  • type: float
  • example: 65.55

metadata.total_silence

Total duration of silences in this file in seconds.

  • type: float
  • example: 34.32

Silence is defined as a period of no speech lasting for 5 seconds or longer.

metadata.silence_percentage

The share of silences in the audio file. This is the total silence duration divided by the total audio duration and is. A high silence ratio can be an indicator for problems.

  • type: float
  • example: 0.18

metadata.word_count

Total number of words.

  • type: integer

metadata.talk_speed

The average number of words per second.

  • type: float
  • example: 3.27

metadata.num_speakers

The number of speakers in the conversation.

metadata.sentiment_score

The overall (aggregated) sentiment in the conversation, using a Contexta NLP model. If missing, no sentiment was detected on the conversation level.

  • type: string
  • possible values: [“negative”, "positive"]

This field is only present if an overall sentiment on conversation level is detected.

metadata.overtalk_duration

The total number of seconds in which speech is detected by both speakers. Only works for dual channel files.

  • type: float

metadata.sentiment_pos_count

The number of positive segments in this conversation.

  • type: integer

This field is only present if there is at least 1 positive segment.

metadata.sentiment_neg_count

The number of negative segments in this conversation.

  • type: integer

This field is only present if there is at least 1 negative segment.

Speaker metrics

These are speaker based performance metrics. These can be found in the SpeakerList section of the transcription for each individual speaker.

[speaker].metadata.keywords

This field contains the most important keywords or phrases from the conversation, using a Contexta NLP model.

  • type: array

example: { [ "compare energy online", "last year", "variable costs", "get a new energy contract", "what's the price" ] }

[speaker].metadata.total_duration

Total duration in seconds.

  • type: float
  • example: 22.95

[speaker].metadata.word_count

Total number of words spoken by this speaker.

  • type: integer
  • example: 77

[speaker].metadata.talk_speed

The talking speed of this speaker, measured in average number of words per second.

  • type: float
  • example: 3.35

[speaker].metadata.talk_percentage

The part of the conversation in which this speaker is talking.

  • type: float
  • example: 0.46

[speaker].metadata.sentiment_score

The overall (aggregated) sentiment in the conversation, using a Contexta NLP model.

  • type: string
  • possible values: [“negative”, "positive"]

[speaker].metadata.sentiment_pos_count

The number of positive segments in this conversation.

  • type: integer

This field is only present if there is at least 1 positive segment.

[speaker].metadata.sentiment_neg_count

The number of negative segments in this conversation.

  • type: integer

This field is only present if there is at least 1 negative segment.

Segment meta data

These are meta data that apply to a specific segment. These can be found in the SegmentList section of the transcription.

[segment].question

A boolean setting that indicates whether the segment is a question, using a Contexta NLP model.

  • type: string
  • possible values: [“True” , “False”]

[segment].metadata.sentiment

Indicates what the sentiment of this segment was, using a Contexta NLP model.

  • type: integer
  • possible values: [-1, 0, 1]

The values correspond to negative, neutral and positive sentiments.