03 / CASE STUDY
NLP / SPEECH AI · 2026
CALLSENSE
An NLP call-intelligence prototype that transcribes audio, summarizes conversations, classifies emotion, and extracts urgency and action items.
CUSTOMER My delivery was delayed again.
AGENT I will raise an escalation.
01 / THESIS
A CALL
LEAVES
MORE THAN
AUDIO.
CallSense starts with conversation and keeps the transformation visible: audio becomes transcript, semantic signals, and a summary that can carry action information forward.
The project treats call intelligence as an interpretation problem, making emotion, urgency, and action extraction readable instead of leaving them buried in the recording.
02 / THE SYSTEM
CONVERSATION BECOMES MEANING.
CallSense keeps the path from call to transcript, semantic signals, and actionable output visible.
03 / CONVERSATION TRACE
THE TRANSCRIPT HOLDS THE SIGNAL.
Speaker turns become inspectable semantic outputs: emotion, urgency, and the action information carried by the conversation.
04 / DECISIONS
MEANING SHOULD REMAIN INSPECTABLE.
Transcript before inference
The conversation is transcribed before semantic signals such as emotion, urgency, and action can be inspected.
TRANSCRIPT / SIGNAL / ACTIONSignals stay explicit
Emotion and urgency remain named outputs rather than being collapsed into an unexplained overall score.
TRANSCRIPT / SIGNAL / ACTIONSummary carries action
The final summary keeps the actionable implication of the conversation visible alongside its interpretation.
TRANSCRIPT / SIGNAL / ACTION05 / SIGNATURE
TRANSCRIPT INTO MEANING.
CUSTOMER My delivery was delayed again.
AGENT I will raise an escalation.
- EMOTION
- ANGER
- URGENCY
- MEDIUM
- ACTION
- ESCALATE + CONFIRM
06 / DEMONSTRATES
CONVERSATION INTO SIGNAL, SUMMARY, AND ACTION.
TRANSCRIBED CONVERSATION
SEMANTIC SIGNALS
ACTION INFORMATION