test

run_flax_speech_recognition_seq2seq_streaming.py

@@ -767,11 +767,13 @@ def main():
         # Add a folder for the mp3 files
         if not os.path.exists(os.path.join(predictions_folder_name,"mp3")):
             os.makedirs(os.path.join(predictions_folder_name,"mp3"))
-         
-        for idx,mp3array in enumerate(eval_samples[0:data_args.number_write_predictions]):
-            audio_segment = AudioSegment(data=mp3array.tobytes(),sample_width=mp3array.dtype.itemsize,frame_rate=16000,channels=1)
-            audio_segment.export(os.path.join(predictions_folder_name, "mp3")+"/pred_"+str(idx)+".mp3", format="mp3")
-            
+        
+        for idx, mp3array in enumerate(eval_samples[0:data_args.number_write_predictions]):
+            audio_segment = AudioSegment(data=mp3array.tobytes(), sample_width=mp3array.dtype.itemsize, frame_rate=16000, channels=1)
+            file_path = os.path.join(predictions_folder_name, "mp3") + "/pred_" + str(idx) + ".mp3"
+            if not os.path.exists(file_path):
+                audio_segment.export(file_path, format="mp3")
+        
         logger.info(f"Created {stats_file_name} and updated the headers of the other stats files")
     
     # 9. Save feature extractor, tokenizer and config

run_streaming_ncc_short.sh

@@ -1,3 +1,4 @@
+export TOKENIZERS_PARALELLISM=true
 python run_flax_speech_recognition_seq2seq_streaming.py \
             --model_name_or_path openai/whisper-tiny \
             --dataset_name NbAiLab/NCC_S3_nrk \