Extracting insights

Creating an Extraction job is the best way to extract insights from your transcripts.

As you already know, Wordcab features four different extraction processes: Emotion Detection, Topic Change Detection, Question/Answer Detection, and Speaker Talk Ratios.

Each of these processes can be used at the same time or individually.

Emotion Detection

Emotion Detection is the process of detecting the emotion of an utterance.

A list of all the emotions Wordcab can detect is available on the Extraction API Concepts page.

To start an Emotion Detection job, you need to send a POST request to the /extract endpoint (see Start Extract) with the following pipeline:

{
  "pipeline": "emotions"
}

🤩 Example of emotions detected in a transcript:

    ...
    "transcript": [
        {
            "text": "This looks great! We can follow-up next week.",
            "speaker": "A",
            "detected_emotion": "joy"
        },
        {
            "text": "Oh wow!",
            "speaker": "B",
            "detected_emotion": "surprise"
        },
    ...

Topic Change Detection

Topic Change Detection is the process of detecting when a topic changes in a conversation.

To start a Topic Change Detection job, you need to send a POST request to the /extract endpoint (see Start Extract) with the following pipeline:

{
  "pipeline": "topic_segments"
}

In the output transcript of that kind of pipeline, 0 indicates that the topic has not changed, and 1 indicates that the topic has changed.

⛱️ Example of a topic change in a transcript:

    ...
    "transcript": [
        {
            "text": "This looks great! We can follow-up next week.",
            "speaker": "A",
            "topic_change": 0
        },
        {
            "text": "On another note, how is your side-project?",
            "speaker": "B",
            "topic_change": 1
        },
    ...

Question/Answer Detection

Question/Answer Detection is the process of detecting when a question is asked and when an answer to that question is given through the conversation.

To start a Question/Answer Detection job, you need to send a POST request to the /extract endpoint (see Start Extract) with the following pipeline:

{
  "pipeline": "questions_answers"
}

🍦 Example of a question/answer in a transcript:

    ...
    "questions_answers": [
        {
            "answer": "I'm writing about a subject that I love, ice cream.",
            "question": "What subject are you writing about?",
            "answer_speaker": "A",
            "question_speaker": "B"
        },
        {
            "answer": "I am definitely a fan of ice cream. I could eat it all day long.",
            "question": "Are you a fan of ice cream?",
            "answer_speaker": "A",
            "question_speaker": "C"
        }
    ],
    ...

Speaker Talk Ratios

Speaker Talk Ratios is the process of detecting the ratio of time each speaker talks in a conversation.

To start a Speaker Talk Ratios job, you need to send a POST request to the /extract endpoint (see Start Extract) with the following pipeline:

{
  "pipeline": "speaker_talk_ratios"
}

🗣️ Example of speaker talk ratios in a transcript:

    ...
    "speaker_talk_ratios": {
        "A": "52.28%",
        "B": "25.33%",
        "C": "22.39%"
    }
    ...

Multiple Extraction Processes

You can use multiple extraction processes at the same time. To do so, you need to send a POST request to the /extract endpoint (see Start Extract) with the following pipeline:

{
  "pipeline": "emotions,topic_segments,questions_answers,speaker_talk_ratios"
}

Each extraction process will be applied to the transcript, and the results will be added to the output. The order you specify the extraction processes in the pipeline does not matter.


What’s Next

⏩ Let's make the Webhook Portal handy.