File size: 24,103 Bytes
c7e434a
60be371
 
c7e434a
 
 
60be371
 
c7e434a
 
 
60be371
9abd5a8
c7e434a
 
60be371
 
c7e434a
60be371
 
 
82e47b6
60be371
 
6dc05e5
 
 
 
 
60be371
 
6dc05e5
 
 
60be371
 
 
 
 
 
 
 
82e47b6
60be371
 
 
9abd5a8
60be371
 
82e47b6
60be371
c7e434a
60be371
 
 
c7e434a
60be371
 
c7e434a
60be371
 
 
c7e434a
60be371
 
 
 
 
c7e434a
60be371
 
c7e434a
60be371
 
 
 
 
 
 
c7e434a
 
60be371
 
c7e434a
60be371
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
9abd5a8
 
60be371
 
 
 
 
 
 
 
9abd5a8
60be371
 
 
 
9abd5a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60be371
9abd5a8
 
 
60be371
9a815f6
 
 
 
60be371
 
 
 
 
9a815f6
9abd5a8
60be371
c7e434a
60be371
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
c7e434a
 
60be371
c7e434a
60be371
c7e434a
 
 
60be371
 
c7e434a
60be371
 
 
 
 
c7e434a
60be371
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
 
30d0bb7
 
 
 
60be371
30d0bb7
 
 
60be371
30d0bb7
 
 
60be371
30d0bb7
 
 
60be371
30d0bb7
60be371
30d0bb7
60be371
30d0bb7
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9abd5a8
 
c7e434a
60be371
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9abd5a8
 
 
60be371
9abd5a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
60be371
 
9abd5a8
 
 
 
 
 
 
 
60be371
 
9abd5a8
 
 
 
 
 
60be371
9abd5a8
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c7e434a
60be371
9abd5a8
60be371
c7e434a
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
"""
Unified Articulation Analysis Service
Gabungan PER-based (dengan reference) dan Clarity-based (tanpa reference)
"""

import torch
import torchaudio
import librosa
import numpy as np
from typing import Dict, List, Tuple, Optional
import re
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
from rapidfuzz import fuzz


class ArticulationService:
    """Analisis artikulasi unified (dengan/tanpa reference text)"""
    
    def __init__(self):
        """Initialize Wav2Vec2 untuk phoneme recognition"""
        print("πŸ—£οΈ Initializing Articulation Service...")
        
        # Load Wav2Vec2 Indonesian model untuk phoneme detection
        model_name = "indonesian-nlp/wav2vec2-indonesian-javanese-sundanese"
        
        # Set cache directory (production: /.cache, local: default)
        import os
        cache_dir = os.environ.get('HF_HOME', '/.cache')
        
        try:
            print(f"πŸ“¦ Loading Wav2Vec2 model: {model_name}")
            print(f"πŸ“ Cache directory: {cache_dir}")
            self.processor = Wav2Vec2Processor.from_pretrained(model_name, cache_dir=cache_dir)
            self.model = Wav2Vec2ForCTC.from_pretrained(model_name, cache_dir=cache_dir)
            self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
            self.model.to(self.device)
            self.model_loaded = True
            print(f"πŸ’» Device: {self.device}")
        except Exception as e:
            print(f"⚠️  Warning: Failed to load Wav2Vec2 model: {e}")
            print("⚠️  Will use fallback articulation analysis")
            self.model_loaded = False
        
        # Filler words bahasa Indonesia
        self.filler_words = [
            'eh', 'ehm', 'em', 'aa', 'ah', 'mm', 'hmm',
            'anu', 'itu', 'gitu', 'kayak', 'seperti',
            'ya', 'yaa', 'nah', 'terus', 'jadi', 'soalnya'
        ]
        
        print("βœ… Articulation Service ready!\n")
    
    def extract_audio_features(self, audio_path: str) -> Tuple[Dict, torch.Tensor, int]:
        """Extract fitur audio untuk analisis artikulasi"""
        print(f"🎡 Extracting audio features from: {audio_path}")
        
        # Load audio
        waveform, sr = torchaudio.load(audio_path)
        
        # Convert to mono
        if waveform.shape[0] > 1:
            waveform = torch.mean(waveform, dim=0, keepdim=True)
        
        # Resample ke 16kHz jika perlu
        if sr != 16000:
            resampler = torchaudio.transforms.Resample(sr, 16000)
            waveform = resampler(waveform)
            sr = 16000
        
        # Convert to numpy
        audio = waveform.squeeze().numpy()
        
        # Extract features
        features = {
            'duration': len(audio) / sr,
            'rms_energy': np.sqrt(np.mean(audio**2)),
            'zero_crossing_rate': librosa.zero_crossings(audio).sum() / len(audio),
            'spectral_centroid': np.mean(librosa.feature.spectral_centroid(y=audio, sr=sr)),
            'spectral_rolloff': np.mean(librosa.feature.spectral_rolloff(y=audio, sr=sr))
        }
        
        print(f"   Duration: {features['duration']:.2f}s")
        print(f"   RMS Energy: {features['rms_energy']:.4f}")
        
        return features, waveform, sr
    
    def analyze_phoneme_clarity(self, waveform: torch.Tensor, sr: int) -> Dict:
        """Analisis kejelasan phoneme menggunakan Wav2Vec2"""
        print("πŸ” Analyzing phoneme clarity...")
        
        if self.model is None or self.processor is None:
            print("⚠️  Wav2Vec2 not available, using fallback")
            return {
                'clarity_score': 70.0,  # Default score
                'avg_confidence': 0.7,
                'min_confidence': 0.5,
                'confidence_std': 0.15,
                'consistency': 0.85
            }
        
        try:
            # Prepare input
            inputs = self.processor(
                waveform.squeeze().numpy(),
                sampling_rate=sr,
                return_tensors="pt",
                padding=True
            )
            
            inputs = {k: v.to(self.device) for k, v in inputs.items()}
            
            # Get logits
            with torch.no_grad():
                logits = self.model(**inputs).logits
            
            # Get confidence scores
            probs = torch.nn.functional.softmax(logits, dim=-1)
            max_probs = torch.max(probs, dim=-1).values
            
            # Calculate clarity metrics
            avg_confidence = torch.mean(max_probs).item()
            min_confidence = torch.min(max_probs).item()
            confidence_std = torch.std(max_probs).item()
            
            # Clarity score (0-100)
            clarity_score = avg_confidence * 100
            
            print(f"   Clarity Score: {clarity_score:.2f}%")
            print(f"   Avg Confidence: {avg_confidence:.3f}")
            
            return {
                'clarity_score': clarity_score,
                'avg_confidence': avg_confidence,
                'min_confidence': min_confidence,
                'confidence_std': confidence_std,
                'consistency': 1 - confidence_std
            }
        except Exception as e:
            print(f"⚠️  Error in phoneme clarity analysis: {e}")
            return {
                'clarity_score': 70.0,
                'avg_confidence': 0.7,
                'min_confidence': 0.5,
                'confidence_std': 0.15,
                'consistency': 0.85
            }
    
    def detect_filler_words(self, transcript: str) -> Dict:
        """Deteksi kata-kata pengisi (filler words)"""
        print("πŸ”Ž Detecting filler words...")
        
        # Split by whitespace to preserve original form
        words = transcript.split()
        total_words = len(words)
        
        if total_words == 0:
            return {
                'filler_count': 0,
                'filler_words_found': []
            }
        
        # Count filler words using fuzzy matching + exact match for short words
        filler_found = []
        filler_count = 0
        
        for word in words:
            # Clean word for checking (lowercase, remove punctuation)
            clean_word = re.sub(r'[^\w\s]', '', word.lower())
            
            # Skip empty words
            if not clean_word:
                continue
            
            is_filler = False
            
            # For short words (2-3 chars), use exact match to avoid false positives
            if len(clean_word) <= 3:
                if clean_word in self.filler_words:
                    is_filler = True
            else:
                # For longer words, use fuzzy matching with 90% threshold
                for filler_word in self.filler_words:
                    similarity = fuzz.ratio(clean_word, filler_word)
                    if similarity >= 90:  # 90% threshold untuk presisi lebih tinggi
                        is_filler = True
                        break
            
            if is_filler:
                filler_count += 1
                # Keep original word form (with punctuation like 'ehm...')
                if word not in filler_found:
                    filler_found.append(word)
        
        # Calculate filler ratio
        filler_ratio = filler_count / total_words if total_words > 0 else 0
        
        print(f"   Filler Words: {filler_count}/{total_words} ({filler_ratio*100:.1f}%)")
        if filler_found:
            print(f"   Found: {', '.join(filler_found)}")
        
        return {
            'filler_count': filler_count,
            'filler_ratio': filler_ratio,
            'filler_words_found': filler_found
        }
    
    def analyze_speech_rate_stability(self, audio_path: str) -> Dict:
        """Analisis kestabilan kecepatan bicara"""
        print("πŸ“Š Analyzing speech rate stability...")
        
        try:
            # Load audio
            y, sr = librosa.load(audio_path, sr=16000)
            
            # Detect onsets (segment boundaries)
            onset_frames = librosa.onset.onset_detect(y=y, sr=sr, units='frames')
            onset_times = librosa.frames_to_time(onset_frames, sr=sr)
            
            if len(onset_times) < 2:
                print("   ⚠️  Not enough onsets detected")
                return {
                    'stability_score': 50.0,
                    'avg_syllable_rate': 0,
                    'rate_std': 0
                }
            
            # Calculate inter-onset intervals (IOI)
            ioi = np.diff(onset_times)
            
            # Speech rate metrics
            avg_rate = 1 / np.mean(ioi) if len(ioi) > 0 else 0
            rate_std = np.std(ioi) if len(ioi) > 0 else 0
            
            # Stability score (semakin rendah std, semakin stabil)
            stability_score = max(0, 100 - (rate_std * 100))
            
            print(f"   Stability Score: {stability_score:.2f}%")
            print(f"   Syllable Rate: {avg_rate:.2f}/s")
            
            return {
                'stability_score': stability_score,
                'avg_syllable_rate': avg_rate,
                'rate_std': rate_std,
                'onset_count': len(onset_times)
            }
        except Exception as e:
            print(f"⚠️  Error in stability analysis: {e}")
            return {
                'stability_score': 60.0,
                'avg_syllable_rate': 0,
                'rate_std': 0
            }
    
    def calculate_per(self, reference: str, hypothesis: str) -> float:
        """
        Calculate Phoneme Error Rate (word-level approximation)
        Using Levenshtein distance
        """
        ref_words = reference.lower().split()
        hyp_words = hypothesis.lower().split()
        m, n = len(ref_words), len(hyp_words)
        
        # Dynamic programming for edit distance
        dp = [[0] * (n + 1) for _ in range(m + 1)]
        
        for i in range(m + 1):
            dp[i][0] = i
        for j in range(n + 1):
            dp[0][j] = j
        
        for i in range(1, m + 1):
            for j in range(1, n + 1):
                if ref_words[i-1] == hyp_words[j-1]:
                    dp[i][j] = dp[i-1][j-1]
                else:
                    dp[i][j] = 1 + min(
                        dp[i-1][j],    # deletion
                        dp[i][j-1],    # insertion
                        dp[i-1][j-1]   # substitution
                    )
        
        errors = dp[m][n]
        per = (errors / m * 100) if m > 0 else 0
        return per
    
    def calculate_overall_score(
        self,
        clarity: Dict,
        filler: Dict,
        stability: Dict,
        features: Dict,
        per: Optional[float] = None
    ) -> Dict:
        """Hitung skor keseluruhan artikulasi"""
        print("\n🎯 Calculating overall articulation score...")
        
        # Clarity score (0-100)
        clarity_score = clarity['clarity_score']
        
        # Filler score (0-100, semakin sedikit filler semakin baik)
        filler_score = max(0, 100 - (filler['filler_ratio'] * 200))
        
        # Stability score (0-100)
        stability_score = stability['stability_score']
        
        # Energy score (normalized RMS energy)
        energy_score = min(100, features['rms_energy'] * 1000)
        
        if per is not None:
            # Mode 1: WITH REFERENCE - PER based
            # Weight untuk dengan reference
            weights = {
                'per': 0.4,          # 40% - PER adalah gold standard
                'clarity': 0.3,      # 30% - Kejelasan phoneme
                'stability': 0.2,    # 20% - Kestabilan tempo
                'energy': 0.1        # 10% - Energi/volume bicara
            }
            
            # PER score (lower is better, invert to 0-100 scale)
            per_score = max(0, 100 - per)
            
            # Weighted average
            total_score = (
                per_score * weights['per'] +
                clarity_score * weights['clarity'] +
                stability_score * weights['stability'] +
                energy_score * weights['energy']
            )
            
            # Convert to 1-5 scale
            score_5 = int(np.clip(total_score / 20, 1, 5))
            
            # Category
            if score_5 >= 5:
                category = "Sempurna"
                reason = f"PER sangat rendah ({per:.1f}%), artikulasi sangat jelas"
            elif score_5 >= 4:
                category = "Baik"
                reason = f"PER rendah ({per:.1f}%), artikulasi jelas"
            elif score_5 >= 3:
                category = "Cukup"
                reason = f"PER sedang ({per:.1f}%), artikulasi cukup jelas"
            elif score_5 >= 2:
                category = "Kurang"
                reason = f"PER tinggi ({per:.1f}%), banyak kesalahan pengucapan"
            else:
                category = "Buruk"
                reason = f"PER sangat tinggi ({per:.1f}%), artikulasi tidak jelas"
            
            print(f"\nπŸ“Š Score Breakdown (WITH REFERENCE):")
            print(f"   PER:       {per:.1f}% β†’ Score: {per_score:.1f}% (weight: {weights['per']*100:.0f}%)")
            print(f"   Clarity:   {clarity_score:.1f}% (weight: {weights['clarity']*100:.0f}%)")
            print(f"   Stability: {stability_score:.1f}% (weight: {weights['stability']*100:.0f}%)")
            print(f"   Energy:    {energy_score:.1f}% (weight: {weights['energy']*100:.0f}%)")
            print(f"   TOTAL:     {total_score:.1f}% β†’ {score_5}/5")
            
            return {
                'score': score_5,
                'category': category,
                'reason': reason,
                'mode': 'with_reference',
                'details': {
                    'per': round(per, 2),
                    'per_score': round(per_score, 2),
                    'clarity_score': round(clarity_score, 2),
                    'stability_score': round(stability_score, 2),
                    'energy_score': round(energy_score, 2),
                    'total_score': round(total_score, 2)
                }
            }
        else:
            # Mode 2: WITHOUT REFERENCE - Clarity based
            # Weight untuk tanpa reference (TANPA filler component)
            weights = {
                'clarity': 0.5,      # 50% - Kejelasan phoneme paling penting
                'stability': 0.3,    # 30% - Kestabilan tempo
                'energy': 0.2        # 20% - Energi/volume bicara
            }
            
            # Weighted average
            total_score = (
                clarity_score * weights['clarity'] +
                stability_score * weights['stability'] +
                energy_score * weights['energy']
            )
            
            # Convert to 1-5 scale based on percentage ranges
            # 81-100% = 5, 61-80% = 4, 41-60% = 3, 21-40% = 2, 0-20% = 1
            if total_score >= 81:
                score_5 = 5
                category = "Sempurna"
                reason = f"Artikulasi sangat jelas ({total_score:.1f}%) dan konsisten"
            elif total_score >= 61:
                score_5 = 4
                category = "Baik"
                reason = f"Artikulasi jelas ({total_score:.1f}%) dengan tempo stabil"
            elif total_score >= 41:
                score_5 = 3
                category = "Cukup"
                reason = f"Artikulasi cukup jelas ({total_score:.1f}%), ada sedikit variasi tempo"
            elif total_score >= 21:
                score_5 = 2
                category = "Kurang"
                reason = f"Artikulasi kurang jelas ({total_score:.1f}%), tempo tidak stabil"
            else:
                score_5 = 1
                category = "Buruk"
                reason = f"Artikulasi tidak jelas ({total_score:.1f}%) dan sulit dipahami"
            
            print(f"\nπŸ“Š Score Breakdown (WITHOUT REFERENCE):")
            print(f"   Clarity:   {clarity_score:.1f}% (weight: {weights['clarity']*100:.0f}%)")
            print(f"   Stability: {stability_score:.1f}% (weight: {weights['stability']*100:.0f}%)")
            print(f"   Energy:    {energy_score:.1f}% (weight: {weights['energy']*100:.0f}%)")
            print(f"   TOTAL:     {total_score:.1f}% β†’ {score_5}/5")
            
            return {
                'score': score_5,
                'category': category,
                'reason': reason,
                'mode': 'without_reference',
                'details': {
                    'clarity_score': round(clarity_score, 2),
                    'stability_score': round(stability_score, 2),
                    'energy_score': round(energy_score, 2),
                    'total_score': round(total_score, 2)
                }
            }
    
    def analyze(self, audio_path: str, transcript: str, reference_text: Optional[str] = None) -> Dict:
        """
        Analisis artikulasi unified (auto-detect mode)
        
        Args:
            audio_path: Path ke file audio
            transcript: Hasil transcription
            reference_text: Text reference (optional, jika ada gunakan PER mode)
            
        Returns:
            Dict hasil analisis artikulasi
        """
        print("\n" + "="*60)
        if reference_text and reference_text.strip():
            print("πŸ—£οΈ  ARTICULATION ANALYSIS (WITH REFERENCE)")
            mode_desc = "PER-based"
        else:
            print("πŸ—£οΈ  ARTICULATION ANALYSIS (WITHOUT REFERENCE)")
            mode_desc = "Clarity-based"
        print("="*60)
        
        # Extract audio features
        features, waveform, sr = self.extract_audio_features(audio_path)
        
        # Analyze phoneme clarity
        clarity = self.analyze_phoneme_clarity(waveform, sr)
        
        # Detect filler words
        filler = self.detect_filler_words(transcript)
        
        # Analyze speech rate stability
        stability = self.analyze_speech_rate_stability(audio_path)
        
        # Calculate PER if reference provided
        per = None
        if reference_text and reference_text.strip():
            print(f"\nπŸ“ Calculating PER...")
            per = self.calculate_per(reference_text, transcript)
            print(f"   PER: {per:.2f}%")
        
        # Calculate overall score
        result = self.calculate_overall_score(clarity, filler, stability, features, per)
        
        # Add detailed metrics
        result['clarity_metrics'] = {
            'avg_confidence': round(clarity['avg_confidence'], 3),
            'consistency': round(clarity['consistency'], 3)
        }
        
        result['filler_count'] = filler['filler_count']
        result['filler_words'] = filler['filler_words_found']
        
        result['stability_metrics'] = {
            'syllable_rate': round(stability['avg_syllable_rate'], 2),
            'rate_variation': round(stability['rate_std'], 3)
        }
        
        if per is not None:
            result['metrics'] = {
                'reference_words': len(reference_text.split()),
                'transcript_words': len(transcript.split()),
                'per': round(per, 2)
            }
        
        print("\nβœ… Articulation analysis complete!")
        print("="*60 + "\n")
        
        return result
\
\

class ProfanityDetector:
    """Deteksi kata tidak senonoh menggunakan hybrid approach (exact + fuzzy + pattern)"""
    
    # Base profanity words (kata dasar)
    PROFANITY_WORDS = {
        'anjir', 'anjay', 'njir', 'njay', 'anjrit', 'njrit', 'anjim', 'anjing', 
        'anjrot', 'asu', 'babi', 'bacot', 'bajingan', 'banci', 'bangke', 'bangor', 
        'bangsat', 'bego', 'bejad', 'bencong', 'bodat', 'bodoh', 'bugil', 'bundir', 
        'bunuh', 'burik', 'burit', 'cawek', 'cemen', 'cipok', 'cium', 'colai', 'coli', 
        'colmek', 'cukimai', 'cukimay', 'culun', 'cumbu', 'dancuk', 'dewasa', 'dick', 
        'dildo', 'encuk', 'fuck', 'gay', 'gei', 'gembel', 'gey', 'gigolo', 'gila', 
        'goblog', 'goblok', 'haram', 'hencet', 'hentai', 'idiot', 'jablai', 'jablay', 
        'jancok', 'jancuk', 'jangkik', 'jembut', 'jilat', 'jingan', 'kampang', 
        'keparat', 'kimak', 'kirik', 'klentit', 'klitoris', 'konthol', 'kontol', 
        'koplok', 'kunyuk', 'kutang', 'kutis', 'kwontol', 'lonte', 'maho', 
        'masturbasi', 'matane', 'mati', 'memek', 'mesum', 'modar', 'modyar', 'mokad', 
        'najis', 'nazi', 'ndhasmu', 'nenen', 'ngentot', 'ngolom', 'ngulum', 'nigga', 
        'nigger', 'onani', 'oon', 'orgasme', 'paksa', 'pantat', 'pantek', 'pecun', 
        'peli', 'penis', 'pentil', 'pepek', 'perek', 'perkosa', 'piatu', 'porno', 
        'pukimak', 'qontol', 'selangkang', 'sempak', 'senggama', 'setan', 'setubuh', 
        'shit', 'silet', 'silit', 'sinting', 'sodomi', 'stres', 'telanjang', 'telaso', 
        'tete', 'tewas', 'titit', 'togel', 'toket', 'tolol', 'tusbol', 'urin', 'vagina'
    }
    
    # Multi-word profanity phrases
    PROFANITY_PHRASES = {
        'gak ada otak', 'tidak ada otak', 'ga ada otak'
    }
    
    # Character substitution map (leet speak)
    CHAR_SUBSTITUTIONS = {
        '0': 'o', '1': 'i', '3': 'e', '4': 'a', '5': 's', 
        '7': 't', '8': 'b', '@': 'a', '$': 's', '*': ''
    }
    
    @classmethod
    def normalize_word(cls, word: str) -> str:
        """Normalize word by replacing common character substitutions"""
        normalized = word.lower()
        for char, replacement in cls.CHAR_SUBSTITUTIONS.items():
            normalized = normalized.replace(char, replacement)
        return normalized
    
    @classmethod
    def detect_profanity(cls, text: str) -> dict:
        """
        Detect profanity using hybrid approach:
        1. Exact match for quick detection
        2. Fuzzy match for typo variations
        3. Pattern matching for character substitution (leet speak)
        """
        text_lower = text.lower()
        
        # Extract words and normalize
        raw_words = re.findall(r'\w+', text_lower)
        
        found_profanity = []
        profanity_count = 0
        
        # Step 1: Check multi-word phrases first
        for phrase in cls.PROFANITY_PHRASES:
            if phrase in text_lower:
                profanity_count += 1
                if phrase not in found_profanity:
                    found_profanity.append(phrase)
        
        # Step 2: Check individual words
        for word in raw_words:
            is_profane = False
            matched_word = word
            
            # A. Exact match (fastest)
            if word in cls.PROFANITY_WORDS:
                is_profane = True
            
            # B. Normalize and check (handle leet speak: t0l0l β†’ tolol)
            elif len(word) > 0:
                normalized = cls.normalize_word(word)
                if normalized in cls.PROFANITY_WORDS:
                    is_profane = True
                    matched_word = normalized
            
            # C. Fuzzy match for typo variations (anjiir, anjiirr, etc.)
            if not is_profane and len(word) > 3:
                for profane_word in cls.PROFANITY_WORDS:
                    # Only compare words with similar length (Β±3 chars)
                    if abs(len(word) - len(profane_word)) <= 3:
                        similarity = fuzz.ratio(word, profane_word)
                        if similarity >= 85:  # 85% threshold for profanity
                            is_profane = True
                            matched_word = profane_word
                            break
            
            if is_profane:
                profanity_count += 1
                # Keep original word if not already in list
                if word not in found_profanity:
                    found_profanity.append(word)
        
        return {
            'has_profanity': len(found_profanity) > 0,
            'profanity_count': profanity_count,
            'profanity_words': list(set(found_profanity))
        }