Spaces:
Sleeping
Sleeping
| import requests | |
| import base64 | |
| import json | |
| import sys | |
| import glob | |
| import random | |
| import numpy as np | |
| print("=" * 60) | |
| print("PHASE 6: BEHAVIORAL ACCURACY CHECK") | |
| print("=" * 60) | |
| API_URL = "http://localhost:8001/api/voice-detection" | |
| API_KEY = "secret123" | |
| # Get test files (WAV only to avoid Windows file cleanup issues) | |
| human_files = [] | |
| ai_files = [] | |
| for lang in ["tamil", "english", "hindi", "malayalam", "telugu"]: | |
| human_lang = glob.glob(f"dataset/human/{lang}/*.wav") | |
| ai_lang = glob.glob(f"dataset/ai_generated/{lang}/*.wav") # WAV files only | |
| human_files.extend(human_lang) | |
| ai_files.extend(ai_lang) | |
| random.seed(42) | |
| # Sample 5 from each | |
| human_sample = random.sample(human_files, min(5, len(human_files))) | |
| ai_sample = random.sample(ai_files, min(5, len(ai_files))) | |
| print(f"Testing {len(human_sample)} human + {len(ai_sample)} AI samples") | |
| def encode_audio_file(file_path): | |
| with open(file_path, 'rb') as f: | |
| return base64.b64encode(f.read()).decode('utf-8') | |
| def test_audio(file_path, expected_class, test_num): | |
| audio_base64 = encode_audio_file(file_path) | |
| payload = { | |
| "language": "Tamil", # Language is language-agnostic | |
| "audioFormat": "wav", | |
| "audioBase64": audio_base64 | |
| } | |
| headers = { | |
| "x-api-key": API_KEY, | |
| "Content-Type": "application/json" | |
| } | |
| try: | |
| response = requests.post(API_URL, json=payload, headers=headers, timeout=30) | |
| if response.status_code != 200: | |
| error_msg = f"{test_num:2d}. FAIL: Status {response.status_code}\n" | |
| print(error_msg.strip()) | |
| with open("api_errors.txt", "a", encoding="utf-8") as err_f: | |
| err_f.write(error_msg) | |
| return None | |
| data = response.json() | |
| is_correct = (data['classification'] == expected_class) | |
| filename = file_path.split('\\')[-1][:35] | |
| status = "✅" if is_correct else "❌" | |
| print(f"{test_num:2d}. {filename:35s} | {data['classification']:12s} | " | |
| f"Conf: {data['confidenceScore']:.4f} | {status}") | |
| return { | |
| 'file': filename, | |
| 'expected': expected_class, | |
| 'predicted': data['classification'], | |
| 'confidence': data['confidenceScore'], | |
| 'explanation': data['explanation'], | |
| 'correct': is_correct | |
| } | |
| except Exception as e: | |
| error_msg = f"{test_num:2d}. ERROR: {str(e)[:50]}\n" | |
| print(error_msg.strip()) | |
| with open("api_errors.txt", "a", encoding="utf-8") as err_f: | |
| err_f.write(error_msg) | |
| return None | |
| # Run tests | |
| print("\n" + "-" * 80) | |
| print("HUMAN SAMPLES (Expected: HUMAN)") | |
| print("-" * 80) | |
| results = [] | |
| test_num = 1 | |
| for file_path in human_sample: | |
| result = test_audio(file_path, 'HUMAN', test_num) | |
| if result: | |
| results.append(result) | |
| test_num += 1 | |
| print("\n" + "-" * 80) | |
| print("AI-GENERATED SAMPLES (Expected: AI_GENERATED)") | |
| print("-" * 80) | |
| for file_path in ai_sample: | |
| result = test_audio(file_path, 'AI_GENERATED', test_num) | |
| if result: | |
| results.append(result) | |
| test_num += 1 | |
| # Analysis | |
| print("\n" + "=" * 60) | |
| print("ANALYSIS") | |
| print("=" * 60) | |
| if len(results) == 0: | |
| print("❌ No results to analyze") | |
| sys.exit(1) | |
| correct = sum(1 for r in results if r['correct']) | |
| total = len(results) | |
| accuracy = (correct / total * 100) if total > 0 else 0 | |
| print(f"Accuracy: {correct}/{total} ({accuracy:.1f}%)") | |
| # Check explanation variance | |
| explanations = [r['explanation'] for r in results] | |
| unique_explanations = len(set(explanations)) | |
| print(f"Unique explanations: {unique_explanations}/{total}") | |
| if unique_explanations == 1: | |
| print("⚠️ All explanations are identical - may indicate lack of feature diversity") | |
| elif unique_explanations < total * 0.3: | |
| print("⚠️ Low explanation variance") | |
| else: | |
| print("✅ Good explanation variance") | |
| # Check confidence correlation | |
| human_confidences = [r['confidence'] for r in results if r['expected'] == 'HUMAN'] | |
| ai_confidences = [r['confidence'] for r in results if r['expected'] == 'AI_GENERATED'] | |
| if human_confidences: | |
| print(f"Human confidence range: {min(human_confidences):.4f} - {max(human_confidences):.4f}") | |
| if ai_confidences: | |
| print(f"AI confidence range: {min(ai_confidences):.4f} - {max(ai_confidences):.4f}") | |
| # Check confidence is in valid range [0.05, 0.95] | |
| out_of_range = [r for r in results if not (0.05 <= r['confidence'] <= 0.95)] | |
| if out_of_range: | |
| print(f"⚠️ {len(out_of_range)} confidence scores out of range [0.05, 0.95]") | |
| # Final verdict | |
| print("\n" + "=" * 60) | |
| print("VERDICT") | |
| print("=" * 60) | |
| if accuracy >= 80: | |
| if total >= 8: | |
| msg = f"✅ PHASE 6 PASSED: {correct}/{total} correct ({accuracy:.1f}% >= 80%)" | |
| print(msg) | |
| else: | |
| msg = f"⚠️ PHASE 6 PASSED but tested fewer than 10 samples ({total}/10)" | |
| print(msg) | |
| with open("final_report.txt", "w", encoding="utf-8") as rep: | |
| rep.write(msg + f"\nAccuracy: {accuracy:.1f}% ({correct}/{total})") | |
| else: | |
| msg = f"❌ PHASE 6 FAILED: {correct}/{total} correct ({accuracy:.1f}% < 80%)" | |
| print(msg) | |
| with open("final_report.txt", "w", encoding="utf-8") as rep: | |
| rep.write(msg) | |
| sys.exit(1) | |