QA-BOT: Real-Time Voice Agent Analysis 🎙️

Description

QA-BOT is an AI-powered solution for monitoring and evaluating customer service and BPO agent interactions in real time. It processes audio data, analyzes sentiment, detects profanity, evaluates tonality, and provides actionable insights to improve service quality.

---

🚀 Features

Real-Time Transcription – Fast & accurate speech-to-text conversion with timestamps.
Speaker Diarization – Identifies & labels different speakers in a conversation.
Sentiment Analysis – Detects emotional states & conversation context.
Profanity Detection – Automatically flags inappropriate language.
Pause Detection – Analyzes response times & conversation flow.
Tonality Checking – Evaluates voice tone & speech patterns.
Knowledge Accuracy (Upcoming Feature) – Assesses information correctness.

---

🛠 Tech Stack

Component	Technology Used
Transcription	VOSK ASR + Kaldi Recognizer
Sentiment Analysis	NLTK's Sentiment Intensity Analyzer (VADER)
Profanity Detection	Better Profanity Library
Audio Processing	PyAudio, Pydub
Machine Learning (Future)	Hugging Face Transformers

---

🔧 Technical Implementation

1️⃣ Transcription 📝

Key Aspects: High-speed real-time transcription, accuracy, timestamping.

• Approach:
  • Captures live audio via PyAudio.
  • Segments audio into frames for processing.
  • Uses VOSK + Kaldi Recognizer for real-time parsing.
  • Outputs structured transcripts with timestamps.

2️⃣ Speaker Diarization 🎭

Key Aspects: Multi-speaker detection, agent identification.

• Approach:
  • Detects speaker change using pause-based analysis.
  • Converts sentences into embeddings for voice pattern comparison.
  • Identifies agents based on a dictionary of common phrases.

3️⃣ Sentiment Analysis 😊😡

Key Aspects: Understanding emotional tone, detecting negativity.

• Approach:
  • Converts speech to text.
  • Uses VADER Lexicon (NLTK) for sentiment polarity.
  • Labels speech as Positive, Negative, Neutral, Angry, or Sarcastic.

4️⃣ Profanity Detection 🚨

Key Aspects: Detects inappropriate language, raises alerts.

• Approach:
  • Uses Better Profanity Library for real-time monitoring.
  • Flags offensive words in transcripts.

5️⃣ Tonality Analysis 🔊

Key Aspects: Evaluates speech tone, conversational style.

• Approach:
  • Uses pre-trained NLP models (e.g., Hugging Face Transformers).
  • Categorizes tone as Formal, Casual, Assertive, or Urgent.

---

📌 Output Examples

📝 Real-Time Transcription Output

📝 Audio File-Input Transcription Output

📝 Audio File-Input Transcription Output

---

📥 Setup & Installation

🔹 Prerequisites

Ensure you have Python 3.8+ installed.
Install dependencies using:

pip install pyaudio vosk nltk better_profanity transformers torch sentencepiece