Conversation Analyzer

Updated on 19 Jul 2024
3 Minutes to read
Contributors

Article summary

Did you find this summary helpful?

Thank you for your feedback

Analyzes conversation speech. This node detects:

The level of tension between the parties involved
Interruptions in each other's speech
Speaking rates (number of words per second)
Periods of silence when both parties remain quiet
Hesitations
Sentiment of the speech text
Speech overlaps

Inputs

Events:

name	description
SR Milestone	Most of the fields examined by the Conversation Analyzer are provided by this event. Features such as silence durations, speech rates, interrupts, and other characteristics are obtained by compiling channel-based events from the sr-milestone
Emotion	The information from the emotion service is compiled to be displayed in the output based on segment and channel. For example, it can be provided that the agent channel exhibits "angry" sentiment between 4-8 seconds.
Sentiment	The results from the sentiment service are analyzed on an agent-customer basis, and the sentiment information of each segment is reflected in the outcome.

Audio:
none

Outputs

name	subname	description
agent		The analysis belonging to the agent's section is located under the "agent" node.
	anger	The sections in the conversation where the agent speaks continuously without interruptions are referred to as "block." In an example output, the total count of blocks, the count of blocks per minute, and the "start-end time" and "duration" information for each block are included.The emotion state of the audio detected from the channel is expressed. The results can appear as "angry, normal, etc." The "anger" field displayed in the output is shown with specific time intervals.
	blocks	The sections in the conversation where the agent speaks continuously without interruptions are referred to as "block." In an example output, the total count of blocks, the count of blocks per minute, and the "start-end time" and "duration" information for each block are included.
	hesitations	This section displays the "pause" information specific to the agent, along with the time intervals
	interrupt	The sections where the agent interrupts the customer's speech are included in this area of the output.
	speed_letter_per_second	The information obtained by dividing the word count from the channel by the total speech duration is included in this section.
	sr	In this section, the "speech recognition" analysis results derived from the audio obtained from the "agent" channel of the dialogue are displayed. The results include information such as time, text, and duration.
customer		The analysis belonging to the customer's section is located under the "customer" node.
	anger	The sections in the conversation where the customer speaks continuously without interruptions are referred to as "block." In an example output, the total count of blocks, the count of blocks per minute, and the "start-end time" and "duration" information for each block are included.The emotion state of the audio detected from the channel is expressed. The results can appear as "angry, normal, etc." The "anger" field displayed in the output is shown with specific time intervals.
	blocks	The sections in the conversation where the customer speaks continuously without interruptions are referred to as "block." In an example output, the total count of blocks, the count of blocks per minute, and the "start-end time" and "duration" information for each block are included.
	hesitations	This section displays the "pause" information specific to the customer, along with the time intervals
	interrupt	The sections where the customer interrupts the agent's speech are included in this area of the output.
	speed_letter_per_second	The information obtained by dividing the word count from the channel by the total speech duration is included in this section.
	sr	In this section, the "speech recognition" analysis results derived from the audio obtained from the "customer" channel of the dialogue are displayed. The results include information such as time, text, and duration.
overlap		The sections in which both channels speak simultaneously throughout the dialogue are listed in the "overlap" field. The overlap information consists of start time, end time, and duration.
silence		The sections in which no audio is received from both channels throughout the dialogue are listed in the "silence" field. The silence information consists of start time, end time, and duration.

Audio:
none

Remarks :
To obtain data from the Conversation Analyzer, it is necessary to configure it along with the events listed in the "Input" field.

License :
none

Project Structure

Conversation Analyzer is probably the most complex node that requires the output from several nodes working together. The project structure can be seen in the default project ca-offline. The node requires the agent and customer to be separated into two channels. If a mono audio input is given, Speaker Diarizer Node is utilized to do the separation. Then, the appropriate audio segments are filtered and used for Gender, Emotion, Sentiment analysis.

Audio Segment Picker and Flush Barrier are utilized, so that the most appropriate segment (5000-10000 milliseconds) is picked for Language Identification before the transcripts are generated in the SR Http node.

Supported flow types: Batch

Was this article helpful?

What's Next

External Relay