second commit, minor fixings

src/extract_features.py

@@ -34,22 +34,23 @@ model.summary()
 buckets = utils.build_buckets(MAX_SEC, BUCKET_STEP, FRAME_STEP)
 
 
-for foldername in sorted(glob.glob(args.input_dir+'/*')):
-	print(foldername)
+for foldername in sorted(glob.glob(args.input_dir+'/**', recursive = True)):
+
 	#session_name=foldername.split('/')[-1]
 	#print(session_name)
 	#embs=[]
 	#for audio_file in sorted(glob.glob(foldername+'/*/*.flac')): ## parametrizzare il tipo di file (oppure cercare semplicemente tutti i fiels
-	audio_file=foldername ###This is so fucked up... I need to fix
-
-	print('Extracting features from file %s'%audio_file)
-	signal = utils.get_fft_spectrum(audio_file, buckets,SAMPLE_RATE,NUM_FFT,FRAME_LEN,FRAME_STEP,PREEMPHASIS_ALPHA)
-	#embs.append((audio_file.split('.')[0],np.squeeze(model.predict(signal.reshape(1,*signal.shape,1)))))
-	embs = np.squeeze(model.predict(signal.reshape(1,*signal.shape,1)))
-	print(embs.shape)
-	print('Dumping vggvox embeddings')
+	if os.path.isfile(foldername) and foldername.lower().endswith(('.flac', '.wav')):
+		audio_file=foldername ###This is so fucked up... I need to fix
+		#print(audio_file)
+		#print('Extracting features from file %s'%audio_file)
+		signal = utils.get_fft_spectrum(audio_file, buckets,SAMPLE_RATE,NUM_FFT,FRAME_LEN,FRAME_STEP,PREEMPHASIS_ALPHA)
+
+		embs = np.squeeze(model.predict(signal.reshape(1,*signal.shape,1)))
+		#print(embs.shape)
+		#print('Dumping vggvox embeddings')
 	
-	fdump=open(args.output_dir+'/'+audio_file.split('/')[-1].split('.')[0]+'.pkl','wb')
-	pkl.dump(embs, fdump, protocol=pkl.HIGHEST_PROTOCOL)
-	fdump.close()
+		fdump=open(args.output_dir+'/'+audio_file.split('/')[-1].split('.')[0]+'.pkl','wb')
+		pkl.dump(embs, fdump, protocol=pkl.HIGHEST_PROTOCOL)
+		fdump.close()