Batch-wise Speech Transcription

238     for b in range(begin_idx,end_idx):
239         
240         timestamp_xls_filename = os.path.join(output_dir,'transcribe_timestamp_{}_{:02d}.xlsx'.format('_'.join(suffix),b+1))
241         timestamp_writer = pd.ExcelWriter(timestamp_xls_filename)
242         
243         ''' check batch xlsx status 1. complete task 2. proccessing yet finished 3. new ''' 
244         titles = []
245         if os.path.isfile(timestamp_xls_filename) and not(args.reset):
246             print('load {} ...'.format(timestamp_xls_filename))
247             
248             book = load_workbook(timestamp_xls_filename)
249             
250             for ws in book.worksheets:
251                 mmm = re.match(r'(.+)\.wav_(.+)', ws.title)
252                 titles.append(os.path.join(input_dir,mmm.group(1)+ '.wav'))
253             titles = sorted(list(set(titles)))
254             print('number of sheets: {}'.format(len(titles)))
255             if len(titles) == batch_size or (b == int(np.ceil(len(entireSpeechFileList)/batch_size)) and len(titles) == len(entireSpeechFileList) % batch_size):
256                 continue
257             else:           
258                 book = load_workbook(timestamp_xls_filename)
259                 timestamp_writer.book = book
260                 timestamp_writer.sheets = dict((ws.title, ws) for ws in book.worksheets)
261                 speechFileQueue = sorted(list(set(entireSpeechFileList[b*batch_size:(b+1)*batch_size]) - set(titles)))
262 
263                 batch_transcribe_speechFile(speechFileQueue,existSpeechFileList,choices_list,timestamp_writer,args.verbose,args.nr,beta=args.beta)
264         else:
265             print('create new {} !!!'.format(timestamp_xls_filename))
266             speechFileQueue = entireSpeechFileList[b*batch_size:(b+1)*batch_size]
267             
268                 batch_transcribe_speechFile(speechFileQueue,existSpeechFileList,choices_list,timestamp_writer,args.verbose,args.nr,beta=args.beta)

行238-269是一個for loop，範圍range(begin_idx,end_idx)是從begin_idx到end_idx-1。

行240-241是產生每個batch有時間資訊的辨識xlsx檔名timestamp_xls_filename跟檔案物件timestamp_writer。

行244呈現建立一個空的列表titles，用來儲存xlsx裡面已辨識音頻檔案的名稱，來核對每個batch xlsx檔的執行情況。

每個batch xlsx檔有三種執行情況$\colon$

1. batch裡面每個音頻都辨識完成 (行255-256)
2. batch裡面有些音頻辨識完成，有些還沒 (行257-263)
3. batch還沒開始辨識 (行264-268)

行245檢查timestamp_xls_filename檔是否存在跟是否要覆寫，如果為False，代表batch要重頭開始辨識(情況3)，如果檔案存在，再檢查檔案裡面已辨識音頻檔的數量，如果跟batch_size相同或者最後一個batch且數量與最後一個batch的音頻數量相同(行255)，代表該batch為情況1，不然為情況2。

情況1，程式會跳過這個batch，執行下一個batch。 情況2，程式會找出還沒辨識的音頻檔(行261)，呼叫函數batch_transcribe_speechFile()辨識還沒辨識的音頻檔。 情況3，程式會從頭開始辨識batch裡的每一個音頻檔。