Commit
Β·
eb2cf07
1
Parent(s):
1a3e21e
Update
Browse files- README.md +161 -0
- analysis_results.csv +2 -0
- api.py +563 -0
- app.py +229 -0
- main.py +60 -0
- render.yaml +9 -0
- requirements.txt +12 -0
- utils.py +18 -0
- voice_sentiment.py +128 -0
README.md
ADDED
@@ -0,0 +1,161 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Voice Sentiment Analysis System
|
2 |
+
|
3 |
+
<div align="center">
|
4 |
+
<img src="gradio_interface.jpg" alt="Voice Sentiment Analysis Banner" width="100%">
|
5 |
+
</div>
|
6 |
+
|
7 |
+
## Project Description
|
8 |
+
|
9 |
+
This project is an automated solution for analyzing customer satisfaction from voice calls. Built using state-of-the-art machine learning models, it combines **Wav2Vec 2.0** for speech-to-text transcription with **BERT** for sentiment analysis to provide real-time feedback into customer emotions and satisfaction levels.
|
10 |
+
|
11 |
+
### Key Features
|
12 |
+
- **Automatic Speech Recognition**: Convert voice calls to text using Wav2Vec 2.0
|
13 |
+
- **Sentiment Analysis**: Analyze emotional tone using multilingual BERT
|
14 |
+
- **Customer Satisfaction Classification**: Categorize calls as Satisfied, Dissatisfied, or Neutral
|
15 |
+
- **Batch Processing**: Handle multiple audio files simultaneously
|
16 |
+
- **Web Interface**: User-friendly [Gradio](https://www.gradio.app/) interface for easy interaction
|
17 |
+
- **CSV Export**: Detailed results export for further analysis and reporting
|
18 |
+
|
19 |
+
## Models Used
|
20 |
+
|
21 |
+
This project uses pre-trained models hosted on Hugging Face Hub:
|
22 |
+
|
23 |
+
### Speech Recognition Model
|
24 |
+
**Wav2Vec 2.0 - English**
|
25 |
+
- **Model:** `facebook/wav2vec2-large-960h-lv60-self`
|
26 |
+
- **Link:** [https://huggingface.co/facebook/wav2vec2-large-960h-lv60-self](https://huggingface.co/facebook/wav2vec2-large-960h-lv60-self)
|
27 |
+
- **Description:** Large Wav2Vec 2.0 model trained on 960 hours of English LibriSpeech data
|
28 |
+
- **Use:** Audio-to-text transcription
|
29 |
+
|
30 |
+
### Sentiment Analysis Model
|
31 |
+
**BERT - Multilingual Sentiment**
|
32 |
+
- **Model:** `nlptown/bert-base-multilingual-uncased-sentiment`
|
33 |
+
- **Link:** [https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment](https://huggingface.co/nlptown/bert-base-multilingual-uncased-sentiment)
|
34 |
+
- **Description:** Multilingual BERT model fine-tuned for sentiment analysis (1-5 stars)
|
35 |
+
- **Use:** Text sentiment classification
|
36 |
+
|
37 |
+
|
38 |
+
## Project Structure
|
39 |
+
|
40 |
+
```
|
41 |
+
voice-sentiment-project/
|
42 |
+
βββ requirements.txt # Dependencies
|
43 |
+
βββ voice_sentiment.py # Core analyzer class
|
44 |
+
βββ api.py # REST API Server
|
45 |
+
βββ app.py # Gradio web interface
|
46 |
+
βββ main.py # CLI interface
|
47 |
+
βββ utils.py # Utility functions and CSS styling
|
48 |
+
βββ audios/ # Your audio files
|
49 |
+
β βββ call1.wav
|
50 |
+
β βββ call2.mp3
|
51 |
+
β βββ ...
|
52 |
+
βββ analysis_results.csv # Generated results
|
53 |
+
```
|
54 |
+
## Language Support
|
55 |
+
|
56 |
+
### Current Model: English Only
|
57 |
+
This system is currently configured with an English-only Wav2Vec 2.0 model (`facebook/wav2vec2-large-960h-lv60-self`) for optimal English speech recognition performance.
|
58 |
+
|
59 |
+
### For Other Languages
|
60 |
+
To use this system with other languages, you need to change the Wav2Vec 2.0 model in `voice_sentiment.py`.
|
61 |
+
|
62 |
+
## Quick Installation
|
63 |
+
|
64 |
+
```bash
|
65 |
+
pip install -r requirements.txt
|
66 |
+
```
|
67 |
+
|
68 |
+
## Usage
|
69 |
+
|
70 |
+
### 1. Web Interface (Recommended)
|
71 |
+
|
72 |
+
```bash
|
73 |
+
python app.py
|
74 |
+
```
|
75 |
+
|
76 |
+
Opens a web browser interface at `http://localhost:7860`
|
77 |
+
|
78 |
+
### 2. Command Line Interface
|
79 |
+
|
80 |
+
```bash
|
81 |
+
python main.py
|
82 |
+
```
|
83 |
+
|
84 |
+
### 3. Direct Code Usage
|
85 |
+
|
86 |
+
```python
|
87 |
+
from voice_sentiment import VoiceSentimentAnalyzer
|
88 |
+
|
89 |
+
# Initialize
|
90 |
+
analyzer = VoiceSentimentAnalyzer()
|
91 |
+
|
92 |
+
# Analyze one call
|
93 |
+
result = analyzer.analyze_call("call1.wav")
|
94 |
+
print(result)
|
95 |
+
|
96 |
+
# Analyze multiple calls
|
97 |
+
results = analyzer.analyze_batch("audios/")
|
98 |
+
```
|
99 |
+
|
100 |
+
## Example Output
|
101 |
+
|
102 |
+
```python
|
103 |
+
{
|
104 |
+
'file': 'call1.wav',
|
105 |
+
'transcription': 'Hello I am very satisfied with your service',
|
106 |
+
'sentiment': 'POSITIVE',
|
107 |
+
'score': 0.89,
|
108 |
+
'satisfaction': 'Satisfied'
|
109 |
+
}
|
110 |
+
```
|
111 |
+
|
112 |
+
## Simple Workflow
|
113 |
+
|
114 |
+
```
|
115 |
+
Audio File β Transcription (Wav2Vec2) β Sentiment (BERT) β Classification
|
116 |
+
```
|
117 |
+
|
118 |
+
Perfect for analyzing customer call sentiment quickly and easily!
|
119 |
+
|
120 |
+
## Supported Audio Formats
|
121 |
+
|
122 |
+
### **Fully Supported**
|
123 |
+
- **WAV** (.wav) - *Recommended for best quality*
|
124 |
+
- **MP3** (.mp3) - *Most common format*
|
125 |
+
- **M4A** (.m4a) - *Apple audio format*
|
126 |
+
|
127 |
+
### **Audio Specifications**
|
128 |
+
- **Sample Rate**: Automatically converted to 16kHz
|
129 |
+
- **Channels**: Mono or Stereo (converted to mono)
|
130 |
+
- **Duration**: 5 seconds to 10 minutes (optimal: 30 seconds - 2 minutes)
|
131 |
+
- **Quality**: Clear speech, minimal background noise recommended
|
132 |
+
|
133 |
+
### **Not Supported**
|
134 |
+
- Video files (MP4, AVI, MOV, etc.)
|
135 |
+
- Other audio formats (FLAC, OGG, etc.) - *may work but not guaranteed*
|
136 |
+
- Extremely low quality or heavily distorted audio
|
137 |
+
- Files with encryption or DRM protection
|
138 |
+
|
139 |
+
### **Audio Quality Tips**
|
140 |
+
- Use WAV format for highest accuracy
|
141 |
+
- Ensure clear speech recording
|
142 |
+
- Minimize background noise
|
143 |
+
- Optimal recording: 16kHz, 16-bit, mono
|
144 |
+
- Test with short samples first
|
145 |
+
|
146 |
+
## CSV Output & Results
|
147 |
+
|
148 |
+
### **Automatic CSV Generation**
|
149 |
+
When using batch analysis (multiple files), the system automatically generates a detailed CSV file with all results.
|
150 |
+
|
151 |
+
**File**: `analysis_results.csv`
|
152 |
+
|
153 |
+
**Location**: Same folder as the project
|
154 |
+
|
155 |
+
### **CSV Contents**
|
156 |
+
```csv
|
157 |
+
File,Transcription,Sentiment,Score,Satisfaction
|
158 |
+
call1.wav,"Hello I am very satisfied with your service",POSITIVE,0.89,Satisfied
|
159 |
+
call2.wav,"This is unacceptable I want a refund",NEGATIVE,0.92,Dissatisfied
|
160 |
+
call3.wav,"Can you tell me about your pricing",NEUTRAL,0.65,Neutral
|
161 |
+
```
|
analysis_results.csv
ADDED
@@ -0,0 +1,2 @@
|
|
|
|
|
|
|
1 |
+
File,Transcription,Sentiment,Score,Satisfaction
|
2 |
+
satistaction_fr.m4a,HERO GISPE I THEACCUSETRES ATISFE DE VA SERVIC RAMAND GENEREETPAT DE MATRE BONE EFONICE THE MAYOR SE...,POSITIVE,0.34,Neutral
|
api.py
ADDED
@@ -0,0 +1,563 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""
|
3 |
+
REST API for Voice Sentiment Analysis System
|
4 |
+
Provides endpoints for integrating the pipeline into other applications
|
5 |
+
"""
|
6 |
+
|
7 |
+
from flask import Flask, request, jsonify, render_template_string
|
8 |
+
from flask_cors import CORS
|
9 |
+
import os
|
10 |
+
import tempfile
|
11 |
+
import uuid
|
12 |
+
from voice_sentiment import VoiceSentimentAnalyzer
|
13 |
+
import logging
|
14 |
+
|
15 |
+
# Initialize Flask app
|
16 |
+
app = Flask(__name__)
|
17 |
+
CORS(app) # Enable CORS for cross-origin requests
|
18 |
+
|
19 |
+
# Configure logging
|
20 |
+
logging.basicConfig(level=logging.INFO)
|
21 |
+
logger = logging.getLogger(__name__)
|
22 |
+
|
23 |
+
# Initialize the analyzer (singleton)
|
24 |
+
analyzer = None
|
25 |
+
|
26 |
+
def get_analyzer():
|
27 |
+
"""Get or create analyzer instance"""
|
28 |
+
global analyzer
|
29 |
+
if analyzer is None:
|
30 |
+
logger.info("Initializing Voice Sentiment Analyzer...")
|
31 |
+
analyzer = VoiceSentimentAnalyzer()
|
32 |
+
logger.info("Analyzer ready!")
|
33 |
+
return analyzer
|
34 |
+
|
35 |
+
# API Documentation HTML Template
|
36 |
+
API_DOCS_HTML = """
|
37 |
+
<!DOCTYPE html>
|
38 |
+
<html>
|
39 |
+
<head>
|
40 |
+
<title>Voice Sentiment Analysis API Documentation</title>
|
41 |
+
<style>
|
42 |
+
body { font-family: Arial, sans-serif; margin: 40px; line-height: 1.6; }
|
43 |
+
.header { background: #f4f4f4; padding: 20px; border-radius: 5px; margin-bottom: 30px; }
|
44 |
+
.endpoint { background: #f9f9f9; padding: 15px; margin: 20px 0; border-left: 4px solid #007cba; }
|
45 |
+
.method { background: #007cba; color: white; padding: 3px 8px; border-radius: 3px; font-size: 12px; }
|
46 |
+
.method.get { background: #28a745; }
|
47 |
+
.method.post { background: #007cba; }
|
48 |
+
pre { background: #f4f4f4; padding: 15px; border-radius: 5px; overflow-x: auto; }
|
49 |
+
code { background: #f4f4f4; padding: 2px 4px; border-radius: 3px; }
|
50 |
+
.example { margin: 10px 0; }
|
51 |
+
h1 { color: #333; }
|
52 |
+
h2 { color: #007cba; border-bottom: 2px solid #007cba; padding-bottom: 5px; }
|
53 |
+
h3 { color: #555; }
|
54 |
+
</style>
|
55 |
+
</head>
|
56 |
+
<body>
|
57 |
+
<div class="header">
|
58 |
+
<h1>Voice Sentiment Analysis API</h1>
|
59 |
+
<p><strong>Version:</strong> 1.0.0</p>
|
60 |
+
<p><strong>Base URL:</strong> <code>{{ base_url }}</code></p>
|
61 |
+
<p>Analyze customer call sentiment using Wav2Vec 2.0 + BERT pipeline</p>
|
62 |
+
</div>
|
63 |
+
|
64 |
+
<h2>Authentication</h2>
|
65 |
+
<p>No authentication required for this API.</p>
|
66 |
+
|
67 |
+
<h2>Supported Audio Formats</h2>
|
68 |
+
<ul>
|
69 |
+
<li><strong>WAV</strong> (.wav) - Recommended</li>
|
70 |
+
<li><strong>MP3</strong> (.mp3)</li>
|
71 |
+
<li><strong>M4A</strong> (.m4a)</li>
|
72 |
+
</ul>
|
73 |
+
|
74 |
+
<h2>API Endpoints</h2>
|
75 |
+
|
76 |
+
<div class="endpoint">
|
77 |
+
<h3><span class="method get">GET</span> /docs</h3>
|
78 |
+
<p><strong>Description:</strong> This documentation page</p>
|
79 |
+
<p><strong>Response:</strong> HTML documentation</p>
|
80 |
+
</div>
|
81 |
+
|
82 |
+
<div class="endpoint">
|
83 |
+
<h3><span class="method get">GET</span> /health</h3>
|
84 |
+
<p><strong>Description:</strong> Health check endpoint</p>
|
85 |
+
<p><strong>Response:</strong></p>
|
86 |
+
<pre><code>{
|
87 |
+
"status": "healthy",
|
88 |
+
"service": "Voice Sentiment Analysis API",
|
89 |
+
"version": "1.0.0"
|
90 |
+
}</code></pre>
|
91 |
+
</div>
|
92 |
+
|
93 |
+
<div class="endpoint">
|
94 |
+
<h3><span class="method post">POST</span> /analyze</h3>
|
95 |
+
<p><strong>Description:</strong> Analyze a single audio file for sentiment</p>
|
96 |
+
<p><strong>Content-Type:</strong> multipart/form-data</p>
|
97 |
+
<p><strong>Parameters:</strong></p>
|
98 |
+
<ul>
|
99 |
+
<li><code>audio</code> (file, required): Audio file to analyze</li>
|
100 |
+
</ul>
|
101 |
+
|
102 |
+
<div class="example">
|
103 |
+
<p><strong>Example Request (cURL):</strong></p>
|
104 |
+
<pre><code>curl -X POST \\
|
105 |
+
-F "audio=@call1.wav" \\
|
106 |
+
{{ base_url }}/analyze</code></pre>
|
107 |
+
</div>
|
108 |
+
|
109 |
+
<div class="example">
|
110 |
+
<p><strong>Example Response:</strong></p>
|
111 |
+
<pre><code>{
|
112 |
+
"success": true,
|
113 |
+
"data": {
|
114 |
+
"filename": "call1.wav",
|
115 |
+
"transcription": "Hello I am very satisfied with your service",
|
116 |
+
"sentiment": "POSITIVE",
|
117 |
+
"confidence_score": 0.89,
|
118 |
+
"satisfaction": "Satisfied"
|
119 |
+
},
|
120 |
+
"processing_id": "uuid-string"
|
121 |
+
}</code></pre>
|
122 |
+
</div>
|
123 |
+
|
124 |
+
<div class="example">
|
125 |
+
<p><strong>Error Response:</strong></p>
|
126 |
+
<pre><code>{
|
127 |
+
"error": "Unsupported file format",
|
128 |
+
"message": "Supported formats: .wav, .mp3, .m4a, .flac",
|
129 |
+
"received": ".txt"
|
130 |
+
}</code></pre>
|
131 |
+
</div>
|
132 |
+
</div>
|
133 |
+
|
134 |
+
<div class="endpoint">
|
135 |
+
<h3><span class="method post">POST</span> /analyze/batch</h3>
|
136 |
+
<p><strong>Description:</strong> Analyze multiple audio files</p>
|
137 |
+
<p><strong>Content-Type:</strong> multipart/form-data</p>
|
138 |
+
<p><strong>Parameters:</strong></p>
|
139 |
+
<ul>
|
140 |
+
<li><code>audio</code> (files, required): Multiple audio files to analyze</li>
|
141 |
+
</ul>
|
142 |
+
|
143 |
+
<div class="example">
|
144 |
+
<p><strong>Example Request (cURL):</strong></p>
|
145 |
+
<pre><code>curl -X POST \\
|
146 |
+
-F "audio=@call1.wav" \\
|
147 |
+
-F "audio=@call2.mp3" \\
|
148 |
+
{{ base_url }}/analyze/batch</code></pre>
|
149 |
+
</div>
|
150 |
+
|
151 |
+
<div class="example">
|
152 |
+
<p><strong>Example Response:</strong></p>
|
153 |
+
<pre><code>{
|
154 |
+
"success": true,
|
155 |
+
"batch_id": "uuid-string",
|
156 |
+
"statistics": {
|
157 |
+
"total_files": 2,
|
158 |
+
"sentiment_distribution": {
|
159 |
+
"POSITIVE": {"count": 1, "percentage": 50.0},
|
160 |
+
"NEGATIVE": {"count": 1, "percentage": 50.0}
|
161 |
+
},
|
162 |
+
"satisfaction_distribution": {
|
163 |
+
"Satisfied": {"count": 1, "percentage": 50.0},
|
164 |
+
"Dissatisfied": {"count": 1, "percentage": 50.0}
|
165 |
+
}
|
166 |
+
},
|
167 |
+
"results": [
|
168 |
+
{
|
169 |
+
"filename": "call1.wav",
|
170 |
+
"transcription": "Hello I am satisfied",
|
171 |
+
"sentiment": "POSITIVE",
|
172 |
+
"confidence_score": 0.89,
|
173 |
+
"satisfaction": "Satisfied",
|
174 |
+
"success": true
|
175 |
+
},
|
176 |
+
{
|
177 |
+
"filename": "call2.mp3",
|
178 |
+
"transcription": "This is terrible service",
|
179 |
+
"sentiment": "NEGATIVE",
|
180 |
+
"confidence_score": 0.92,
|
181 |
+
"satisfaction": "Dissatisfied",
|
182 |
+
"success": true
|
183 |
+
}
|
184 |
+
],
|
185 |
+
"processed_files": 2,
|
186 |
+
"total_uploaded": 2
|
187 |
+
}</code></pre>
|
188 |
+
</div>
|
189 |
+
</div>
|
190 |
+
|
191 |
+
<div class="endpoint">
|
192 |
+
<h3><span class="method get">GET</span> /models/info</h3>
|
193 |
+
<p><strong>Description:</strong> Get information about loaded models</p>
|
194 |
+
<p><strong>Response:</strong></p>
|
195 |
+
<pre><code>{
|
196 |
+
"speech_recognition": {
|
197 |
+
"model": "facebook/wav2vec2-large-960h-lv60-self",
|
198 |
+
"type": "Wav2Vec 2.0",
|
199 |
+
"language": "English",
|
200 |
+
"description": "Large Wav2Vec 2.0 model for English speech recognition"
|
201 |
+
},
|
202 |
+
"sentiment_analysis": {
|
203 |
+
"model": "nlptown/bert-base-multilingual-uncased-sentiment",
|
204 |
+
"type": "BERT",
|
205 |
+
"language": "Multilingual",
|
206 |
+
"description": "Multilingual BERT for sentiment analysis"
|
207 |
+
},
|
208 |
+
"supported_formats": [".wav", ".mp3", ".m4a", ".flac"],
|
209 |
+
"classifications": {
|
210 |
+
"sentiments": ["POSITIVE", "NEGATIVE", "NEUTRAL"],
|
211 |
+
"satisfaction": ["Satisfied", "Dissatisfied", "Neutral"]
|
212 |
+
}
|
213 |
+
}</code></pre>
|
214 |
+
</div>
|
215 |
+
|
216 |
+
<h2>Response Codes</h2>
|
217 |
+
<ul>
|
218 |
+
<li><strong>200</strong> - Success</li>
|
219 |
+
<li><strong>400</strong> - Bad Request (invalid file, missing parameters)</li>
|
220 |
+
<li><strong>404</strong> - Endpoint Not Found</li>
|
221 |
+
<li><strong>413</strong> - File Too Large (>16MB)</li>
|
222 |
+
<li><strong>500</strong> - Internal Server Error</li>
|
223 |
+
</ul>
|
224 |
+
|
225 |
+
<h2>Integration Examples</h2>
|
226 |
+
|
227 |
+
<h3>Python</h3>
|
228 |
+
<pre><code>import requests
|
229 |
+
|
230 |
+
# Single file analysis
|
231 |
+
with open('audio.wav', 'rb') as f:
|
232 |
+
response = requests.post(
|
233 |
+
'{{ base_url }}/analyze',
|
234 |
+
files={'audio': f}
|
235 |
+
)
|
236 |
+
result = response.json()
|
237 |
+
print(f"Sentiment: {result['data']['sentiment']}")
|
238 |
+
|
239 |
+
# Batch analysis
|
240 |
+
files = [
|
241 |
+
('audio', open('call1.wav', 'rb')),
|
242 |
+
('audio', open('call2.mp3', 'rb'))
|
243 |
+
]
|
244 |
+
response = requests.post('{{ base_url }}/analyze/batch', files=files)
|
245 |
+
result = response.json()
|
246 |
+
print(f"Processed {result['processed_files']} files")</code></pre>
|
247 |
+
|
248 |
+
<h3>JavaScript</h3>
|
249 |
+
<pre><code>// Single file upload
|
250 |
+
const formData = new FormData();
|
251 |
+
formData.append('audio', fileInput.files[0]);
|
252 |
+
|
253 |
+
fetch('{{ base_url }}/analyze', {
|
254 |
+
method: 'POST',
|
255 |
+
body: formData
|
256 |
+
})
|
257 |
+
.then(response => response.json())
|
258 |
+
.then(data => {
|
259 |
+
console.log('Sentiment:', data.data.sentiment);
|
260 |
+
});</code></pre>
|
261 |
+
|
262 |
+
<h3>Node.js</h3>
|
263 |
+
<pre><code>const fs = require('fs');
|
264 |
+
const FormData = require('form-data');
|
265 |
+
|
266 |
+
const form = new FormData();
|
267 |
+
form.append('audio', fs.createReadStream('call.wav'));
|
268 |
+
|
269 |
+
fetch('{{ base_url }}/analyze', {
|
270 |
+
method: 'POST',
|
271 |
+
body: form
|
272 |
+
})
|
273 |
+
.then(response => response.json())
|
274 |
+
.then(data => console.log(data));</code></pre>
|
275 |
+
|
276 |
+
<h2>Rate Limits</h2>
|
277 |
+
<p>Currently no rate limits are enforced. For production use, consider implementing rate limiting.</p>
|
278 |
+
|
279 |
+
<h2>File Size Limits</h2>
|
280 |
+
<ul>
|
281 |
+
<li><strong>Maximum file size:</strong> 16MB per file</li>
|
282 |
+
<li><strong>Recommended:</strong> Keep files under 5MB for faster processing</li>
|
283 |
+
<li><strong>Optimal duration:</strong> 30 seconds to 2 minutes</li>
|
284 |
+
</ul>
|
285 |
+
|
286 |
+
<footer style="margin-top: 50px; padding-top: 20px; border-top: 1px solid #eee; color: #666;">
|
287 |
+
<p>Voice Sentiment Analysis API - Powered by Wav2Vec 2.0 + BERT</p>
|
288 |
+
</footer>
|
289 |
+
</body>
|
290 |
+
</html>
|
291 |
+
"""
|
292 |
+
|
293 |
+
@app.route('/docs', methods=['GET'])
|
294 |
+
@app.route('/documentation', methods=['GET'])
|
295 |
+
@app.route('/', methods=['GET'])
|
296 |
+
def api_documentation():
|
297 |
+
"""API Documentation page"""
|
298 |
+
base_url = request.url_root.rstrip('/')
|
299 |
+
return render_template_string(API_DOCS_HTML, base_url=base_url)
|
300 |
+
|
301 |
+
@app.route('/health', methods=['GET'])
|
302 |
+
def health_check():
|
303 |
+
"""Health check endpoint"""
|
304 |
+
return jsonify({
|
305 |
+
"status": "healthy",
|
306 |
+
"service": "Voice Sentiment Analysis API",
|
307 |
+
"version": "1.0.0"
|
308 |
+
})
|
309 |
+
|
310 |
+
@app.route('/analyze', methods=['POST'])
|
311 |
+
def analyze_audio():
|
312 |
+
"""
|
313 |
+
Analyze a single audio file
|
314 |
+
|
315 |
+
Expected: multipart/form-data with 'audio' file
|
316 |
+
Returns: JSON with analysis results
|
317 |
+
"""
|
318 |
+
try:
|
319 |
+
# Check if file is present
|
320 |
+
if 'audio' not in request.files:
|
321 |
+
return jsonify({
|
322 |
+
"error": "No audio file provided",
|
323 |
+
"message": "Please upload an audio file using the 'audio' field"
|
324 |
+
}), 400
|
325 |
+
|
326 |
+
audio_file = request.files['audio']
|
327 |
+
|
328 |
+
# Check if file is selected
|
329 |
+
if audio_file.filename == '':
|
330 |
+
return jsonify({
|
331 |
+
"error": "No file selected",
|
332 |
+
"message": "Please select an audio file"
|
333 |
+
}), 400
|
334 |
+
|
335 |
+
# Validate file extension
|
336 |
+
allowed_extensions = ['.wav', '.mp3', '.m4a', '.flac']
|
337 |
+
file_ext = os.path.splitext(audio_file.filename)[1].lower()
|
338 |
+
|
339 |
+
if file_ext not in allowed_extensions:
|
340 |
+
return jsonify({
|
341 |
+
"error": "Unsupported file format",
|
342 |
+
"message": f"Supported formats: {', '.join(allowed_extensions)}",
|
343 |
+
"received": file_ext
|
344 |
+
}), 400
|
345 |
+
|
346 |
+
# Save file temporarily
|
347 |
+
temp_id = str(uuid.uuid4())
|
348 |
+
temp_filename = f"temp_audio_{temp_id}{file_ext}"
|
349 |
+
temp_path = os.path.join(tempfile.gettempdir(), temp_filename)
|
350 |
+
|
351 |
+
audio_file.save(temp_path)
|
352 |
+
|
353 |
+
try:
|
354 |
+
# Analyze the audio
|
355 |
+
analyzer = get_analyzer()
|
356 |
+
result = analyzer.analyze_call(temp_path)
|
357 |
+
|
358 |
+
# Clean up temporary file
|
359 |
+
os.remove(temp_path)
|
360 |
+
|
361 |
+
# Return results
|
362 |
+
return jsonify({
|
363 |
+
"success": True,
|
364 |
+
"data": {
|
365 |
+
"filename": audio_file.filename,
|
366 |
+
"transcription": result['transcription'],
|
367 |
+
"sentiment": result['sentiment'],
|
368 |
+
"confidence_score": round(result['score'], 3),
|
369 |
+
"satisfaction": result['satisfaction']
|
370 |
+
},
|
371 |
+
"processing_id": temp_id
|
372 |
+
})
|
373 |
+
|
374 |
+
except Exception as e:
|
375 |
+
# Clean up on error
|
376 |
+
if os.path.exists(temp_path):
|
377 |
+
os.remove(temp_path)
|
378 |
+
raise e
|
379 |
+
|
380 |
+
except Exception as e:
|
381 |
+
logger.error(f"Error processing audio: {str(e)}")
|
382 |
+
return jsonify({
|
383 |
+
"error": "Processing failed",
|
384 |
+
"message": str(e)
|
385 |
+
}), 500
|
386 |
+
|
387 |
+
@app.route('/analyze/batch', methods=['POST'])
|
388 |
+
def analyze_batch():
|
389 |
+
"""
|
390 |
+
Analyze multiple audio files
|
391 |
+
|
392 |
+
Expected: multipart/form-data with multiple 'audio' files
|
393 |
+
Returns: JSON with batch analysis results
|
394 |
+
"""
|
395 |
+
try:
|
396 |
+
# Check if files are present
|
397 |
+
if 'audio' not in request.files:
|
398 |
+
return jsonify({
|
399 |
+
"error": "No audio files provided",
|
400 |
+
"message": "Please upload audio files using the 'audio' field"
|
401 |
+
}), 400
|
402 |
+
|
403 |
+
audio_files = request.files.getlist('audio')
|
404 |
+
|
405 |
+
if not audio_files or all(f.filename == '' for f in audio_files):
|
406 |
+
return jsonify({
|
407 |
+
"error": "No files selected",
|
408 |
+
"message": "Please select audio files"
|
409 |
+
}), 400
|
410 |
+
|
411 |
+
results = []
|
412 |
+
temp_files = []
|
413 |
+
batch_id = str(uuid.uuid4())
|
414 |
+
|
415 |
+
try:
|
416 |
+
# Process each file
|
417 |
+
for i, audio_file in enumerate(audio_files):
|
418 |
+
if audio_file.filename == '':
|
419 |
+
continue
|
420 |
+
|
421 |
+
# Validate file extension
|
422 |
+
allowed_extensions = ['.wav', '.mp3', '.m4a', '.flac']
|
423 |
+
file_ext = os.path.splitext(audio_file.filename)[1].lower()
|
424 |
+
|
425 |
+
if file_ext not in allowed_extensions:
|
426 |
+
results.append({
|
427 |
+
"filename": audio_file.filename,
|
428 |
+
"error": f"Unsupported format: {file_ext}",
|
429 |
+
"success": False
|
430 |
+
})
|
431 |
+
continue
|
432 |
+
|
433 |
+
# Save file temporarily
|
434 |
+
temp_filename = f"batch_{batch_id}_{i}{file_ext}"
|
435 |
+
temp_path = os.path.join(tempfile.gettempdir(), temp_filename)
|
436 |
+
temp_files.append(temp_path)
|
437 |
+
|
438 |
+
audio_file.save(temp_path)
|
439 |
+
|
440 |
+
# Analyze the audio
|
441 |
+
analyzer = get_analyzer()
|
442 |
+
result = analyzer.analyze_call(temp_path)
|
443 |
+
|
444 |
+
results.append({
|
445 |
+
"filename": audio_file.filename,
|
446 |
+
"transcription": result['transcription'],
|
447 |
+
"sentiment": result['sentiment'],
|
448 |
+
"confidence_score": round(result['score'], 3),
|
449 |
+
"satisfaction": result['satisfaction'],
|
450 |
+
"success": True
|
451 |
+
})
|
452 |
+
|
453 |
+
# Calculate statistics
|
454 |
+
successful_results = [r for r in results if r.get('success', False)]
|
455 |
+
total_files = len(successful_results)
|
456 |
+
|
457 |
+
if total_files > 0:
|
458 |
+
sentiment_counts = {}
|
459 |
+
satisfaction_counts = {}
|
460 |
+
|
461 |
+
for result in successful_results:
|
462 |
+
sentiment = result['sentiment']
|
463 |
+
satisfaction = result['satisfaction']
|
464 |
+
|
465 |
+
sentiment_counts[sentiment] = sentiment_counts.get(sentiment, 0) + 1
|
466 |
+
satisfaction_counts[satisfaction] = satisfaction_counts.get(satisfaction, 0) + 1
|
467 |
+
|
468 |
+
statistics = {
|
469 |
+
"total_files": total_files,
|
470 |
+
"sentiment_distribution": {
|
471 |
+
k: {"count": v, "percentage": round(v/total_files*100, 1)}
|
472 |
+
for k, v in sentiment_counts.items()
|
473 |
+
},
|
474 |
+
"satisfaction_distribution": {
|
475 |
+
k: {"count": v, "percentage": round(v/total_files*100, 1)}
|
476 |
+
for k, v in satisfaction_counts.items()
|
477 |
+
}
|
478 |
+
}
|
479 |
+
else:
|
480 |
+
statistics = {"total_files": 0, "message": "No files processed successfully"}
|
481 |
+
|
482 |
+
return jsonify({
|
483 |
+
"success": True,
|
484 |
+
"batch_id": batch_id,
|
485 |
+
"statistics": statistics,
|
486 |
+
"results": results,
|
487 |
+
"processed_files": len(successful_results),
|
488 |
+
"total_uploaded": len([f for f in audio_files if f.filename != ''])
|
489 |
+
})
|
490 |
+
|
491 |
+
finally:
|
492 |
+
# Clean up temporary files
|
493 |
+
for temp_path in temp_files:
|
494 |
+
if os.path.exists(temp_path):
|
495 |
+
os.remove(temp_path)
|
496 |
+
|
497 |
+
except Exception as e:
|
498 |
+
logger.error(f"Error processing batch: {str(e)}")
|
499 |
+
return jsonify({
|
500 |
+
"error": "Batch processing failed",
|
501 |
+
"message": str(e)
|
502 |
+
}), 500
|
503 |
+
|
504 |
+
@app.route('/models/info', methods=['GET'])
|
505 |
+
def model_info():
|
506 |
+
"""Get information about loaded models"""
|
507 |
+
return jsonify({
|
508 |
+
"speech_recognition": {
|
509 |
+
"model": "facebook/wav2vec2-large-960h-lv60-self",
|
510 |
+
"type": "Wav2Vec 2.0",
|
511 |
+
"language": "English",
|
512 |
+
"description": "Large Wav2Vec 2.0 model for English speech recognition"
|
513 |
+
},
|
514 |
+
"sentiment_analysis": {
|
515 |
+
"model": "nlptown/bert-base-multilingual-uncased-sentiment",
|
516 |
+
"type": "BERT",
|
517 |
+
"language": "Multilingual",
|
518 |
+
"description": "Multilingual BERT for sentiment analysis (1-5 stars)"
|
519 |
+
},
|
520 |
+
"supported_formats": [".wav", ".mp3", ".m4a", ".flac"],
|
521 |
+
"classifications": {
|
522 |
+
"sentiments": ["POSITIVE", "NEGATIVE", "NEUTRAL"],
|
523 |
+
"satisfaction": ["Satisfied", "Dissatisfied", "Neutral"]
|
524 |
+
}
|
525 |
+
})
|
526 |
+
|
527 |
+
@app.errorhandler(413)
|
528 |
+
def file_too_large(error):
|
529 |
+
"""Handle file too large error"""
|
530 |
+
return jsonify({
|
531 |
+
"error": "File too large",
|
532 |
+
"message": "Audio file exceeds maximum size limit"
|
533 |
+
}), 413
|
534 |
+
|
535 |
+
@app.errorhandler(404)
|
536 |
+
def not_found(error):
|
537 |
+
"""Handle 404 errors"""
|
538 |
+
return jsonify({
|
539 |
+
"error": "Endpoint not found",
|
540 |
+
"message": "The requested endpoint does not exist",
|
541 |
+
"available_endpoints": [
|
542 |
+
"GET /health - Health check",
|
543 |
+
"POST /analyze - Analyze single audio file",
|
544 |
+
"POST /analyze/batch - Analyze multiple audio files",
|
545 |
+
"GET /models/info - Get model information"
|
546 |
+
]
|
547 |
+
}), 404
|
548 |
+
|
549 |
+
if __name__ == '__main__':
|
550 |
+
# Configuration
|
551 |
+
HOST = os.getenv('API_HOST', '0.0.0.0')
|
552 |
+
PORT = int(os.getenv('API_PORT', 8000))
|
553 |
+
DEBUG = os.getenv('API_DEBUG', 'False').lower() == 'true'
|
554 |
+
|
555 |
+
# Set maximum file size (16MB)
|
556 |
+
app.config['MAX_CONTENT_LENGTH'] = 16 * 1024 * 1024
|
557 |
+
|
558 |
+
print(f"Starting Voice Sentiment Analysis API...")
|
559 |
+
print(f"Server: http://{HOST}:{PORT}")
|
560 |
+
print(f"Health check: http://{HOST}:{PORT}/health")
|
561 |
+
print(f"Documentation: See README for API usage examples")
|
562 |
+
|
563 |
+
app.run(host=HOST, port=PORT, debug=DEBUG)
|
app.py
ADDED
@@ -0,0 +1,229 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""
|
3 |
+
Gradio Interface for Voice Sentiment Analysis
|
4 |
+
Wav2Vec 2.0 + BERT Pipeline
|
5 |
+
"""
|
6 |
+
|
7 |
+
import gradio as gr
|
8 |
+
import pandas as pd
|
9 |
+
import os
|
10 |
+
from utils import custom_css
|
11 |
+
from voice_sentiment import VoiceSentimentAnalyzer
|
12 |
+
|
13 |
+
# Initialize model (once)
|
14 |
+
print("Loading models...")
|
15 |
+
analyzer = VoiceSentimentAnalyzer()
|
16 |
+
print("Models ready!")
|
17 |
+
|
18 |
+
def analyze_audio_file(audio_file):
|
19 |
+
"""Analyze an uploaded audio file"""
|
20 |
+
if audio_file is None:
|
21 |
+
return "No audio file provided", "", "", ""
|
22 |
+
|
23 |
+
try:
|
24 |
+
# Analyze the call
|
25 |
+
result = analyzer.analyze_call(audio_file)
|
26 |
+
|
27 |
+
# Format results
|
28 |
+
transcription = result['transcription']
|
29 |
+
sentiment = result['sentiment']
|
30 |
+
score = f"{result['score']:.2f}"
|
31 |
+
satisfaction = result['satisfaction']
|
32 |
+
|
33 |
+
# Emoji based on sentiment
|
34 |
+
emoji_map = {
|
35 |
+
"POSITIVE": "π",
|
36 |
+
"NEGATIVE": "π ",
|
37 |
+
"NEUTRAL": "π"
|
38 |
+
}
|
39 |
+
emoji = emoji_map.get(sentiment, "β")
|
40 |
+
|
41 |
+
status = f"Analysis completed {emoji}"
|
42 |
+
|
43 |
+
return status, transcription, sentiment, score, satisfaction
|
44 |
+
|
45 |
+
except Exception as e:
|
46 |
+
error_msg = f"Analysis error: {str(e)}"
|
47 |
+
return error_msg, "", "", "", ""
|
48 |
+
|
49 |
+
def analyze_batch_files(files):
|
50 |
+
"""Analyze multiple audio files"""
|
51 |
+
if not files:
|
52 |
+
return "No files provided", None
|
53 |
+
|
54 |
+
try:
|
55 |
+
results = []
|
56 |
+
|
57 |
+
for file in files:
|
58 |
+
result = analyzer.analyze_call(file.name)
|
59 |
+
results.append({
|
60 |
+
"File": os.path.basename(file.name),
|
61 |
+
"Transcription": result['transcription'][:100] + "..." if len(result['transcription']) > 100 else result['transcription'],
|
62 |
+
"Sentiment": result['sentiment'],
|
63 |
+
"Score": round(result['score'], 2),
|
64 |
+
"Satisfaction": result['satisfaction']
|
65 |
+
})
|
66 |
+
|
67 |
+
# Create DataFrame for display
|
68 |
+
df = pd.DataFrame(results)
|
69 |
+
csv_filename = "analysis_results.csv"
|
70 |
+
|
71 |
+
print(f"Saving {len(df)} rows to CSV...")
|
72 |
+
df.to_csv(csv_filename, index=False)
|
73 |
+
print(f"CSV saved successfully") #
|
74 |
+
|
75 |
+
# Verify CSV was created and has content
|
76 |
+
if os.path.exists(csv_filename): # β NEW DEBUG BLOCK
|
77 |
+
file_size = os.path.getsize(csv_filename)
|
78 |
+
print(f"CSV file exists, size: {file_size} bytes")
|
79 |
+
else:
|
80 |
+
print("CSV file was not created!")
|
81 |
+
|
82 |
+
|
83 |
+
# Statistics
|
84 |
+
total = len(results)
|
85 |
+
positive = len([r for r in results if r['Sentiment'] == 'POSITIVE'])
|
86 |
+
negative = len([r for r in results if r['Sentiment'] == 'NEGATIVE'])
|
87 |
+
neutral = len([r for r in results if r['Sentiment'] == 'NEUTRAL'])
|
88 |
+
|
89 |
+
stats = f"""π Statistics:
|
90 |
+
β’ Total: {total} calls
|
91 |
+
β’ Positive: {positive} ({positive/total*100:.1f}%)
|
92 |
+
β’ Negative: {negative} ({negative/total*100:.1f}%)
|
93 |
+
β’ Neutral: {neutral} ({neutral/total*100:.1f}%)"""
|
94 |
+
|
95 |
+
return stats, df
|
96 |
+
|
97 |
+
except Exception as e:
|
98 |
+
error_msg = f"Analysis error: {str(e)}"
|
99 |
+
return error_msg, None
|
100 |
+
|
101 |
+
# Gradio Interface
|
102 |
+
with gr.Blocks(title="Voice Sentiment Analysis", theme=gr.themes.Soft(), css=custom_css) as app:
|
103 |
+
|
104 |
+
gr.Markdown("""
|
105 |
+
# Voice Sentiment Analysis System
|
106 |
+
### Wav2Vec 2.0 + BERT Pipeline
|
107 |
+
|
108 |
+
Automatically analyze customer call sentiment and classify satisfaction.
|
109 |
+
""")
|
110 |
+
|
111 |
+
with gr.Tabs():
|
112 |
+
|
113 |
+
# Tab 1: Single file analysis
|
114 |
+
with gr.Tab("Single File"):
|
115 |
+
gr.Markdown("### Analyze one voice call")
|
116 |
+
|
117 |
+
with gr.Row():
|
118 |
+
with gr.Column():
|
119 |
+
audio_input = gr.Audio(
|
120 |
+
type="filepath",
|
121 |
+
label="Upload your audio file"
|
122 |
+
)
|
123 |
+
|
124 |
+
analyze_btn = gr.Button(
|
125 |
+
"Analyze",
|
126 |
+
variant="primary",
|
127 |
+
size="lg"
|
128 |
+
)
|
129 |
+
|
130 |
+
with gr.Column():
|
131 |
+
status_output = gr.Textbox(
|
132 |
+
label="π Status",
|
133 |
+
interactive=False
|
134 |
+
)
|
135 |
+
|
136 |
+
transcription_output = gr.Textbox(
|
137 |
+
label="π Transcription",
|
138 |
+
lines=3,
|
139 |
+
interactive=False
|
140 |
+
)
|
141 |
+
|
142 |
+
with gr.Row():
|
143 |
+
sentiment_output = gr.Textbox(
|
144 |
+
label="π Sentiment",
|
145 |
+
interactive=False
|
146 |
+
)
|
147 |
+
score_output = gr.Textbox(
|
148 |
+
label="π― Confidence Score",
|
149 |
+
interactive=False
|
150 |
+
)
|
151 |
+
|
152 |
+
satisfaction_output = gr.Textbox(
|
153 |
+
label="π Customer Satisfaction",
|
154 |
+
interactive=False
|
155 |
+
)
|
156 |
+
|
157 |
+
# Tab 2: Multiple files analysis
|
158 |
+
with gr.Tab("Multiple Files"):
|
159 |
+
gr.Markdown("### Analyze multiple calls in batch")
|
160 |
+
|
161 |
+
files_input = gr.File(
|
162 |
+
file_count="multiple",
|
163 |
+
file_types=[".wav", ".mp3", ".m4a"],
|
164 |
+
label="Upload your audio files"
|
165 |
+
)
|
166 |
+
|
167 |
+
batch_analyze_btn = gr.Button(
|
168 |
+
"Analyze All",
|
169 |
+
variant="primary",
|
170 |
+
size="lg"
|
171 |
+
)
|
172 |
+
|
173 |
+
batch_status = gr.Textbox(
|
174 |
+
label="Statistics",
|
175 |
+
lines=6,
|
176 |
+
interactive=False
|
177 |
+
)
|
178 |
+
|
179 |
+
results_table = gr.Dataframe(
|
180 |
+
label="Detailed Results",
|
181 |
+
interactive=False
|
182 |
+
)
|
183 |
+
|
184 |
+
# Tab 3: Information
|
185 |
+
with gr.Tab("Information"):
|
186 |
+
gr.Markdown("""
|
187 |
+
### How it works?
|
188 |
+
|
189 |
+
**3-step pipeline:**
|
190 |
+
1. **Audio β Text**: Transcription with Wav2Vec 2.0
|
191 |
+
2. **Text β Sentiment**: Analysis with multilingual BERT
|
192 |
+
3. **Classification**: Customer satisfaction (Satisfied/Dissatisfied/Neutral)
|
193 |
+
|
194 |
+
### Supported formats
|
195 |
+
- WAV (recommended)
|
196 |
+
- MP3
|
197 |
+
- M4A
|
198 |
+
|
199 |
+
### Classifications
|
200 |
+
- **π Satisfied**: Positive sentiment with high confidence
|
201 |
+
- **π Dissatisfied**: Negative sentiment with high confidence
|
202 |
+
- **π Neutral**: Neutral sentiment or low confidence
|
203 |
+
|
204 |
+
### Tips
|
205 |
+
- Clear audio quality recommended
|
206 |
+
- Optimal duration: 10 seconds to 2 minutes
|
207 |
+
- Avoid excessive background noise
|
208 |
+
""")
|
209 |
+
|
210 |
+
# Event connections
|
211 |
+
analyze_btn.click(
|
212 |
+
fn=analyze_audio_file,
|
213 |
+
inputs=[audio_input],
|
214 |
+
outputs=[status_output, transcription_output, sentiment_output, score_output, satisfaction_output]
|
215 |
+
)
|
216 |
+
|
217 |
+
batch_analyze_btn.click(
|
218 |
+
fn=analyze_batch_files,
|
219 |
+
inputs=[files_input],
|
220 |
+
outputs=[batch_status, results_table]
|
221 |
+
)
|
222 |
+
|
223 |
+
# Launch the application
|
224 |
+
if __name__ == "__main__":
|
225 |
+
app.launch(
|
226 |
+
share=True, # Creates a public link
|
227 |
+
server_name="0.0.0.0", # Accessible from other machines
|
228 |
+
server_port=7860
|
229 |
+
)
|
main.py
ADDED
@@ -0,0 +1,60 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
"""
|
3 |
+
Main script for voice sentiment analysis
|
4 |
+
"""
|
5 |
+
|
6 |
+
from voice_sentiment import VoiceSentimentAnalyzer
|
7 |
+
import os
|
8 |
+
|
9 |
+
def main():
|
10 |
+
"""Main function"""
|
11 |
+
print("VOICE SENTIMENT ANALYSIS SYSTEM")
|
12 |
+
print("="*50)
|
13 |
+
|
14 |
+
# Initialize the system
|
15 |
+
analyzer = VoiceSentimentAnalyzer()
|
16 |
+
|
17 |
+
# Simple menu
|
18 |
+
while True:
|
19 |
+
print("\nOptions:")
|
20 |
+
print("1. Analyze an audio file")
|
21 |
+
print("2. Analyze a folder of calls")
|
22 |
+
print("3. Exit")
|
23 |
+
|
24 |
+
choice = input("\nYour choice (1-3): ").strip()
|
25 |
+
|
26 |
+
if choice == "1":
|
27 |
+
# Single file analysis
|
28 |
+
file_path = input("Audio file path: ").strip()
|
29 |
+
|
30 |
+
if os.path.exists(file_path):
|
31 |
+
try:
|
32 |
+
result = analyzer.analyze_call(file_path)
|
33 |
+
print("\nAnalysis completed!")
|
34 |
+
except Exception as e:
|
35 |
+
print(f"Error: {e}")
|
36 |
+
else:
|
37 |
+
print("File not found!")
|
38 |
+
|
39 |
+
elif choice == "2":
|
40 |
+
# Folder analysis
|
41 |
+
folder_path = input("Folder path: ").strip()
|
42 |
+
|
43 |
+
if os.path.exists(folder_path):
|
44 |
+
try:
|
45 |
+
results = analyzer.analyze_batch(folder_path)
|
46 |
+
print(f"\n{len(results)} files analyzed!")
|
47 |
+
except Exception as e:
|
48 |
+
print(f"Error: {e}")
|
49 |
+
else:
|
50 |
+
print("Folder not found!")
|
51 |
+
|
52 |
+
elif choice == "3":
|
53 |
+
print("Goodbye!")
|
54 |
+
break
|
55 |
+
|
56 |
+
else:
|
57 |
+
print("Invalid choice!")
|
58 |
+
|
59 |
+
if __name__ == "__main__":
|
60 |
+
main()
|
render.yaml
ADDED
@@ -0,0 +1,9 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
services:
|
2 |
+
- type: web
|
3 |
+
name: voice-sentiment-api
|
4 |
+
env: python
|
5 |
+
buildCommand: pip install -r requirements.txt
|
6 |
+
startCommand: gunicorn --bind 0.0.0.0:$PORT --timeout 300 api:app
|
7 |
+
envVars:
|
8 |
+
- key: PYTHON_VERSION
|
9 |
+
value: 3.9.18
|
requirements.txt
ADDED
@@ -0,0 +1,12 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
torch>=1.9.0
|
2 |
+
transformers>=4.20.0
|
3 |
+
librosa>=0.9.0
|
4 |
+
pandas>=1.3.0
|
5 |
+
numpy>=1.21.0
|
6 |
+
scipy>=1.7.0
|
7 |
+
torchaudio>=0.9.0
|
8 |
+
soundfile>=0.10.0
|
9 |
+
gradio>=4.0.0
|
10 |
+
flask>=2.0.0
|
11 |
+
flask-cors>=3.0.0
|
12 |
+
gunicorn>=20.0.0
|
utils.py
ADDED
@@ -0,0 +1,18 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# Custom CSS for Helvetica font
|
2 |
+
custom_css = """
|
3 |
+
* {
|
4 |
+
font-family: "Helvetica Neue", Helvetica, Arial, sans-serif !important;
|
5 |
+
}
|
6 |
+
|
7 |
+
.gradio-container {
|
8 |
+
font-family: "Helvetica Neue", Helvetica, Arial, sans-serif !important;
|
9 |
+
}
|
10 |
+
|
11 |
+
.gr-textbox, .gr-button, .gr-markdown, .gr-label {
|
12 |
+
font-family: "Helvetica Neue", Helvetica, Arial, sans-serif !important;
|
13 |
+
}
|
14 |
+
|
15 |
+
h1, h2, h3, h4, h5, h6 {
|
16 |
+
font-family: "Helvetica Neue", Helvetica, Arial, sans-serif !important;
|
17 |
+
}
|
18 |
+
"""
|
voice_sentiment.py
ADDED
@@ -0,0 +1,128 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
Simple Voice Sentiment Analysis System
|
3 |
+
Wav2Vec 2.0 + BERT Pipeline
|
4 |
+
"""
|
5 |
+
|
6 |
+
import torch
|
7 |
+
import librosa
|
8 |
+
import numpy as np
|
9 |
+
from transformers import (
|
10 |
+
Wav2Vec2ForCTC,
|
11 |
+
Wav2Vec2Tokenizer,
|
12 |
+
pipeline
|
13 |
+
)
|
14 |
+
import pandas as pd
|
15 |
+
import os
|
16 |
+
|
17 |
+
class VoiceSentimentAnalyzer:
|
18 |
+
"""Simple Pipeline: Audio β Transcription β Sentiment Analysis"""
|
19 |
+
|
20 |
+
def __init__(self):
|
21 |
+
print("Loading models...")
|
22 |
+
|
23 |
+
# ASR Model (Speech-to-Text)
|
24 |
+
self.asr_tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
|
25 |
+
self.asr_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-960h-lv60-self")
|
26 |
+
|
27 |
+
# Sentiment Model
|
28 |
+
self.sentiment_analyzer = pipeline(
|
29 |
+
"sentiment-analysis",
|
30 |
+
model="nlptown/bert-base-multilingual-uncased-sentiment"
|
31 |
+
)
|
32 |
+
|
33 |
+
print("Models loaded!")
|
34 |
+
|
35 |
+
def audio_to_text(self, audio_path):
|
36 |
+
"""Convert audio to text"""
|
37 |
+
# Load and preprocess audio
|
38 |
+
audio, sr = librosa.load(audio_path, sr=16000)
|
39 |
+
|
40 |
+
# Transcription with Wav2Vec2
|
41 |
+
input_values = self.asr_tokenizer(audio, return_tensors="pt", sampling_rate=16000).input_values
|
42 |
+
|
43 |
+
with torch.no_grad():
|
44 |
+
logits = self.asr_model(input_values).logits
|
45 |
+
|
46 |
+
predicted_ids = torch.argmax(logits, dim=-1)
|
47 |
+
transcription = self.asr_tokenizer.decode(predicted_ids[0])
|
48 |
+
|
49 |
+
return transcription.strip()
|
50 |
+
|
51 |
+
def text_to_sentiment(self, text):
|
52 |
+
"""Analyze sentiment of the text"""
|
53 |
+
if not text:
|
54 |
+
return {"sentiment": "NEUTRAL", "score": 0.0}
|
55 |
+
|
56 |
+
result = self.sentiment_analyzer(text)[0]
|
57 |
+
|
58 |
+
# Convert labels to simple format
|
59 |
+
label_map = {
|
60 |
+
"1 star": "NEGATIVE", "2 stars": "NEGATIVE",
|
61 |
+
"3 stars": "NEUTRAL",
|
62 |
+
"4 stars": "POSITIVE", "5 stars": "POSITIVE"
|
63 |
+
}
|
64 |
+
|
65 |
+
sentiment = label_map.get(result['label'], result['label'])
|
66 |
+
|
67 |
+
return {
|
68 |
+
"sentiment": sentiment,
|
69 |
+
"score": result['score']
|
70 |
+
}
|
71 |
+
|
72 |
+
def classify_satisfaction(self, sentiment, score):
|
73 |
+
"""Classify customer satisfaction"""
|
74 |
+
if sentiment == "POSITIVE" and score > 0.7:
|
75 |
+
return "Satisfied"
|
76 |
+
elif sentiment == "NEGATIVE" and score > 0.7:
|
77 |
+
return "Dissatisfied"
|
78 |
+
else:
|
79 |
+
return "Neutral"
|
80 |
+
|
81 |
+
def analyze_call(self, audio_path):
|
82 |
+
"""Complete pipeline: Audio β Sentiment β Classification"""
|
83 |
+
print(f"Analyzing: {audio_path}")
|
84 |
+
|
85 |
+
# 1. Audio β Text
|
86 |
+
transcription = self.audio_to_text(audio_path)
|
87 |
+
print(f"Transcription: {transcription}")
|
88 |
+
|
89 |
+
# 2. Text β Sentiment
|
90 |
+
sentiment_result = self.text_to_sentiment(transcription)
|
91 |
+
print(f"Sentiment: {sentiment_result['sentiment']} (score: {sentiment_result['score']:.2f})")
|
92 |
+
|
93 |
+
# 3. Satisfaction classification
|
94 |
+
satisfaction = self.classify_satisfaction(sentiment_result['sentiment'], sentiment_result['score'])
|
95 |
+
print(f"Satisfaction: {satisfaction}")
|
96 |
+
|
97 |
+
return {
|
98 |
+
"file": os.path.basename(audio_path),
|
99 |
+
"transcription": transcription,
|
100 |
+
"sentiment": sentiment_result['sentiment'],
|
101 |
+
"score": sentiment_result['score'],
|
102 |
+
"satisfaction": satisfaction
|
103 |
+
}
|
104 |
+
|
105 |
+
def analyze_batch(self, audio_folder):
|
106 |
+
"""Analyze a folder of calls"""
|
107 |
+
results = []
|
108 |
+
|
109 |
+
for filename in os.listdir(audio_folder):
|
110 |
+
if filename.endswith(('.wav', '.mp3', '.m4a')):
|
111 |
+
audio_path = os.path.join(audio_folder, filename)
|
112 |
+
result = self.analyze_call(audio_path)
|
113 |
+
results.append(result)
|
114 |
+
print("-" * 50)
|
115 |
+
|
116 |
+
# Save to CSV
|
117 |
+
df = pd.DataFrame(results)
|
118 |
+
df.to_csv("analysis_results.csv", index=False)
|
119 |
+
print(f"Results saved: analysis_results.csv")
|
120 |
+
|
121 |
+
# Quick statistics
|
122 |
+
print("\nSTATISTICS:")
|
123 |
+
print(f"Total calls: {len(results)}")
|
124 |
+
sentiment_counts = df['sentiment'].value_counts()
|
125 |
+
for sentiment, count in sentiment_counts.items():
|
126 |
+
print(f"{sentiment}: {count}")
|
127 |
+
|
128 |
+
return df
|