File size: 12,288 Bytes
5d4c1d3
 
7111ecb
5d4c1d3
 
 
 
7111ecb
5d4c1d3
 
 
 
 
 
 
 
 
a6526f0
5d4c1d3
7111ecb
5d4c1d3
 
 
7111ecb
5d4c1d3
 
7111ecb
 
 
 
 
 
 
5d4c1d3
 
 
 
 
 
7111ecb
a6526f0
5d4c1d3
 
 
7111ecb
a6526f0
5d4c1d3
7111ecb
 
 
 
 
5d4c1d3
7111ecb
 
 
5d4c1d3
7111ecb
 
5d4c1d3
7111ecb
5d4c1d3
7111ecb
5d4c1d3
7111ecb
5d4c1d3
7111ecb
 
 
 
 
5d4c1d3
 
7111ecb
a6526f0
5d4c1d3
7111ecb
5d4c1d3
7111ecb
 
5d4c1d3
7111ecb
5d4c1d3
 
7111ecb
a6526f0
5d4c1d3
a6526f0
5d4c1d3
 
a6526f0
5d4c1d3
a6526f0
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
5d4c1d3
7111ecb
 
 
 
 
 
5d4c1d3
7111ecb
5d4c1d3
 
7111ecb
 
 
 
 
 
 
 
 
 
 
 
 
 
5d4c1d3
 
 
7111ecb
5d4c1d3
 
7111ecb
5d4c1d3
 
 
 
a6526f0
5d4c1d3
 
 
7111ecb
5d4c1d3
 
 
7111ecb
5d4c1d3
4701923
 
5d4c1d3
 
 
 
4701923
 
 
5d4c1d3
 
7111ecb
 
 
5d4c1d3
 
 
 
 
 
 
 
 
 
 
7111ecb
 
 
 
5d4c1d3
 
7111ecb
5d4c1d3
7111ecb
5d4c1d3
4701923
7111ecb
5d4c1d3
 
 
 
7111ecb
 
5d4c1d3
 
 
7111ecb
5d4c1d3
 
 
 
 
 
 
 
 
7111ecb
5d4c1d3
 
 
 
 
 
7111ecb
5d4c1d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7111ecb
 
5d4c1d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7111ecb
 
 
 
5d4c1d3
a6526f0
7111ecb
 
 
 
5d4c1d3
 
 
 
 
 
 
7111ecb
5d4c1d3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# =======================================================================
# 1. 匯入區 (Imports)
#    - 新增了 pyopenjtalk 和 MeCab
# =======================================================================
import torch
import soundfile as sf
import librosa
from transformers import Wav2Vec2Processor, HubertForCTC
import os
import pyopenjtalk
import MeCab
import numpy as np
from datetime import datetime, timezone
import re

# =======================================================================
# 2. 全域變數與配置區 (Global Variables & Config)
#    【已修改】移除了全域的 processor 和 model 變數。
# =======================================================================
# 自動檢測可用設備
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
print(f"INFO: ASR_jp_jp.py is configured to use device: {DEVICE}")

# 設定為日語 ASR 模型
MODEL_NAME = "prj-beatrice/japanese-hubert-base-phoneme-ctc-v3"

# 初始化 MeCab 分詞器
# -Owakati 選項能直接輸出以空格分隔的單詞,非常方便
try:
    mecab_tagger = MeCab.Tagger("-Owakati")
except RuntimeError:
    print("ERROR: MeCab Tagger 初始化失敗。請確保 mecab 和 mecab-ipadic-utf8 已正確安裝。")
    mecab_tagger = None

# =======================================================================
# 3. 核心業務邏輯區 (Core Business Logic)
# =======================================================================

# -----------------------------------------------------------------------
# 3.1. 模型載入函數
#      【已刪除】舊的 load_model() 函數已被移除。
# -----------------------------------------------------------------------

# -----------------------------------------------------------------------
# 3.2. 日語 G2P 輔助函數 (此檔案最核心的修改)
#      【保持不變】
# -----------------------------------------------------------------------
def _get_target_phonemes_by_word(text: str) -> tuple[list[str], list[list[str]]]:
    if not mecab_tagger:
        raise RuntimeError("MeCab Tagger 未初始化,無法處理日語文本。")

    words = mecab_tagger.parse(text).strip().split()
    
    target_words_original = []
    target_ipa_by_word = []

    for word in words:
        if not word:
            continue
        
        phonemes_str = pyopenjtalk.g2p(word, kana=False)
        
        cleaned_phonemes = re.sub(r'\s+', ' ', phonemes_str).strip()
        
        phoneme_list = cleaned_phonemes.split()
        
        if word and phoneme_list:
            target_words_original.append(word)
            target_ipa_by_word.append(phoneme_list)
            
    return target_words_original, target_ipa_by_word

# -----------------------------------------------------------------------
# 3.3. 音素切分函數 (用於處理 ASR 的輸出)
#      【保持不變】
# -----------------------------------------------------------------------
def _tokenize_asr_output(phoneme_string: str) -> list:
    """
    將 ASR 模型輸出的音素字串切分為列表。
    此模型的輸出是單字元音素,以空格分隔。
    """
    return phoneme_string.split()

# -----------------------------------------------------------------------
# 3.4. 核心分析函數 (主入口)
#      【已修改】將模型載入和快取邏輯合併至此。
# -----------------------------------------------------------------------
def analyze(audio_file_path: str, target_sentence: str, cache: dict = {}) -> dict:
    """
    接收音訊檔案路徑和目標日語句子,回傳詳細的發音分析字典。
    模型會被載入並儲存在此函數獨立的 'cache' 中,實現狀態隔離。
    """
    # 檢查快取中是否已有模型,如果沒有則載入
    if "model" not in cache:
        print(f"快取未命中 (ASR_jp_jp)。正在載入模型 '{MODEL_NAME}'...")
        try:
            # 載入模型並存入此函數的快取字典
            cache["processor"] = Wav2Vec2Processor.from_pretrained(MODEL_NAME)
            cache["model"] = HubertForCTC.from_pretrained(MODEL_NAME) # <-- 使用 HubertForCTC
            cache["model"].to(DEVICE)
            print(f"模型 '{MODEL_NAME}' 已載入並快取。")
        except Exception as e:
            print(f"處理或載入模型 '{MODEL_NAME}' 時發生錯誤: {e}")
            raise RuntimeError(f"Failed to load model '{MODEL_NAME}': {e}")

    # 從此函數的獨立快取中獲取模型和處理器
    processor = cache["processor"]
    model = cache["model"]

    # --- 以下為原始分析邏輯,保持不變 ---

    # 【關鍵步驟 1: G2P】
    target_words_original, target_ipa_by_word = _get_target_phonemes_by_word(target_sentence)

    if not target_words_original:
        print("警告: G2P 處理後目標句子為空。")
        return _format_to_json_structure([], target_sentence, [])

    # 【關鍵步驟 2: ASR】
    try:
        speech, sample_rate = sf.read(audio_file_path)
        if len(speech) == 0:
            print("警告: 音訊檔案為空。")
            user_ipa_full = ""
        else:
            if sample_rate != 16000:
                speech = librosa.resample(y=speech, orig_sr=sample_rate, target_sr=16000)
            
            input_values = processor(speech, sampling_rate=16000, return_tensors="pt").input_values
            input_values = input_values.to(DEVICE)
            with torch.no_grad():
                logits = model(input_values).logits
            predicted_ids = torch.argmax(logits, dim=-1)
            user_ipa_full = processor.decode(predicted_ids[0])

    except Exception as e:
        raise IOError(f"讀取或處理音訊時發生錯誤: {e}")
    
    # 【關鍵步驟 3: 對齊】
    word_alignments = _get_phoneme_alignments_by_word(user_ipa_full, target_ipa_by_word)

    # 【關鍵步驟 4: 格式化】
    return _format_to_json_structure(word_alignments, target_sentence, target_words_original)

# =======================================================================
# 4. 對齊與格式化函數區 (Alignment & Formatting)
#    【保持不變】
# =======================================================================

# -----------------------------------------------------------------------
# 4.1. 對齊函數 (語言無關)
# -----------------------------------------------------------------------
def _get_phoneme_alignments_by_word(user_phoneme_str, target_words_ipa_tokenized):
    """
    使用動態規劃執行音素對齊。此函數是語言無關的。
    """
    user_phonemes = [char for word in user_phoneme_str.split() for char in word]

    target_phonemes_flat = []
    word_boundaries_indices = [] 
    current_idx = 0
    for word_ipa_tokens in target_words_ipa_tokenized:
        flat_tokens = [char for word in word_ipa_tokens for char in word]
        target_phonemes_flat.extend(flat_tokens)
        current_idx += len(flat_tokens)
        word_boundaries_indices.append(current_idx - 1)

    if not target_phonemes_flat:
        return []

    dp = np.zeros((len(user_phonemes) + 1, len(target_phonemes_flat) + 1))
    for i in range(1, len(user_phonemes) + 1): dp[i][0] = i
    for j in range(1, len(target_phonemes_flat) + 1): dp[0][j] = j
    for i in range(1, len(user_phonemes) + 1):
        for j in range(1, len(target_phonemes_flat) + 1):
            cost = 0 if user_phonemes[i-1] == target_phonemes_flat[j-1] else 1
            dp[i][j] = min(dp[i-1][j] + 1, dp[i][j-1] + 1, dp[i-1][j-1] + cost)

    i, j = len(user_phonemes), len(target_phonemes_flat)
    user_path, target_path = [], []
    while i > 0 or j > 0:
        cost = float('inf')
        if i > 0 and j > 0:
            cost = 0 if user_phonemes[i-1] == target_phonemes_flat[j-1] else 1

        if i > 0 and j > 0 and dp[i][j] == dp[i-1][j-1] + cost:
            user_path.insert(0, user_phonemes[i-1]); target_path.insert(0, target_phonemes_flat[j-1]); i -= 1; j -= 1
        elif i > 0 and (j == 0 or dp[i][j] == dp[i-1][j] + 1):
            user_path.insert(0, user_phonemes[i-1]); target_path.insert(0, '-'); i -= 1
        elif j > 0 and (i == 0 or dp[i][j] == dp[i][j-1] + 1):
            user_path.insert(0, '-'); target_path.insert(0, target_phonemes_flat[j-1]); j -= 1
        else:
            break
    
    alignments_by_word = []
    word_start_idx_in_path = 0
    target_phoneme_counter_in_path = 0
    word_boundary_iter = iter(word_boundaries_indices)
    current_word_boundary = next(word_boundary_iter, -1)

    for path_idx, p in enumerate(target_path):
        if p != '-':
            if target_phoneme_counter_in_path == current_word_boundary:
                target_alignment = target_path[word_start_idx_in_path : path_idx + 1]
                user_alignment = user_path[word_start_idx_in_path : path_idx + 1]
                
                alignments_by_word.append({
                    "target": target_alignment,
                    "user": user_alignment
                })
                
                word_start_idx_in_path = path_idx + 1
                current_word_boundary = next(word_boundary_iter, -1)
            
            target_phoneme_counter_in_path += 1
            
    return alignments_by_word

# -----------------------------------------------------------------------
# 4.2. 格式化函數 (語言無關)
# -----------------------------------------------------------------------
def _format_to_json_structure(alignments, sentence, original_words) -> dict:
    """
    將對齊結果格式化為最終的 JSON 結構。此函數是語言無關的。
    """
    total_phonemes = 0
    total_errors = 0
    correct_words_count = 0
    words_data = []

    num_words_to_process = min(len(alignments), len(original_words))

    for i in range(num_words_to_process):
        alignment = alignments[i]
        word_is_correct = True
        phonemes_data = []
        
        min_len = min(len(alignment['target']), len(alignment['user']))
        for j in range(min_len):
            target_phoneme = alignment['target'][j]
            user_phoneme = alignment['user'][j]
            is_match = (user_phoneme == target_phoneme)
            
            phonemes_data.append({
                "target": target_phoneme,
                "user": user_phoneme,
                "isMatch": is_match
            })
            
            if not is_match:
                word_is_correct = False
                if not (user_phoneme == '-' and target_phoneme == '-'):
                    total_errors += 1
        
        if word_is_correct:
            correct_words_count += 1
            
        words_data.append({
            "word": original_words[i],
            "isCorrect": word_is_correct,
            "phonemes": phonemes_data
        })
        
        total_phonemes += sum(1 for p in alignment['target'] if p != '-')

    if len(alignments) < len(original_words):
        for i in range(len(alignments), len(original_words)):
            _, missed_word_ipa_list = _get_target_phonemes_by_word(original_words[i])
            
            phonemes_data = []
            if missed_word_ipa_list:
                for p_ipa in missed_word_ipa_list[0]:
                    phonemes_data.append({"target": p_ipa, "user": "-", "isMatch": False})
                    total_errors += 1
                    total_phonemes += 1

            words_data.append({
                "word": original_words[i],
                "isCorrect": False,
                "phonemes": phonemes_data
            })

    total_words = len(original_words)
    overall_score = (correct_words_count / total_words) * 100 if total_words > 0 else 0
    phoneme_error_rate = (total_errors / total_phonemes) * 100 if total_phonemes > 0 else 0

    final_result = {
        "sentence": sentence,
        "analysisTimestampUTC": datetime.now(timezone.utc).strftime('%Y-%m-%d %H:%M:%S (UTC)'),
        "summary": {
            "overallScore": round(overall_score, 1),
            "totalWords": total_words,
            "correctWords": correct_words_count,
            "phonemeErrorRate": round(phoneme_error_rate, 2),
            "total_errors": total_errors,
            "total_target_phonemes": total_phonemes
        },
        "words": words_data
    }
    
    return final_result