调整默认配置

SpeechModel24.py

@@ -396,8 +396,8 @@ def model(self):
 
 	ms = ModelSpeech(datapath)
 
-	ms.LoadModel(modelpath + 'm24\\speech_model24_e_0_step_411000.model')
-	#ms.TrainModel(datapath, epoch = 50, batch_size = 4, save_step = 500)
+	#ms.LoadModel(modelpath + 'm24/speech_model24_e_0_step_411000.model')
+	ms.TrainModel(datapath, epoch = 50, batch_size = 16, save_step = 500)
 	#ms.TestModel(datapath, str_dataset='test', data_count = 128, out_report = True)
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00241I0053.wav')
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00020I0087.wav')

SpeechModel25.py

@@ -374,7 +374,7 @@ def model(self):
 
 	import tensorflow as tf
 	from keras.backend.tensorflow_backend import set_session
-	os.environ["CUDA_VISIBLE_DEVICES"] = "1"
+	os.environ["CUDA_VISIBLE_DEVICES"] = "0"
 	#进行配置，使用70%的GPU
 	config = tf.ConfigProto()
 	config.gpu_options.per_process_gpu_memory_fraction = 0.93
@@ -403,10 +403,9 @@ def model(self):
 
 	ms = ModelSpeech(datapath)
 
-	# 全场最佳 speech_model25_e_0_step_340000
-	ms.LoadModel(modelpath + 'm25\\speech_model25_e_0_step_545500.model')
-	#ms.TrainModel(datapath, epoch = 50, batch_size = 4, save_step = 500)
-	ms.TestModel(datapath, str_dataset='test', data_count = 128, out_report = True)
+	#ms.LoadModel(modelpath + 'm25/speech_model25_e_0_step_545500.model')
+	ms.TrainModel(datapath, epoch = 50, batch_size = 16, save_step = 500)
+	#ms.TestModel(datapath, str_dataset='test', data_count = 128, out_report = True)
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00241I0053.wav')
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00020I0087.wav')
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\wav\\train\\A11\\A11_167.WAV')

SpeechModel26.py

@@ -420,8 +420,8 @@ def model(self):
 
 	ms = ModelSpeech(datapath)
 
-	#ms.LoadModel(modelpath + 'm26\\speech_model26_e_0_step_221500.model')
-	#ms.TrainModel(datapath, epoch = 50, batch_size = 4, save_step = 1)
+	#ms.LoadModel(modelpath + 'm26/speech_model26_e_0_step_397000.model')
+	ms.TrainModel(datapath, epoch = 50, batch_size = 16, save_step = 500)
 	#ms.TestModel(datapath, str_dataset='test', data_count = 128, out_report = True)
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00241I0053.wav')
 	#r = ms.RecognizeSpeech_FromFile('E:\\语音数据集\\ST-CMDS-20170001_1-OS\\20170001P00020I0087.wav')

test_mspeech.py

@@ -12,11 +12,11 @@
 from keras.backend.tensorflow_backend import set_session
 
 
-from SpeechModel24 import ModelSpeech
+from SpeechModel25 import ModelSpeech
 
 
 os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-#进行配置，使用70%的GPU
+#进行配置，使用90%的GPU
 config = tf.ConfigProto()
 config.gpu_options.per_process_gpu_memory_fraction = 0.9
 #config.gpu_options.allow_growth=True   #不全部占满显存, 按需分配
@@ -44,7 +44,7 @@
 
 ms = ModelSpeech(datapath)
 
-ms.LoadModel(modelpath + 'm24\\speech_model24_e_0_step_42500.model')
+ms.LoadModel(modelpath + 'm25/speech_model25_e_0_step_42500.model')
 
 ms.TestModel(datapath, str_dataset='test', data_count = 128, out_report = True)
 

train_mspeech.py

@@ -12,12 +12,12 @@
 from keras.backend.tensorflow_backend import set_session
 
 
-from SpeechModel24 import ModelSpeech
+from SpeechModel25 import ModelSpeech
 
 os.environ["CUDA_VISIBLE_DEVICES"] = "0"
-#进行配置，使用70%的GPU
+#进行配置，使用95%的GPU
 config = tf.ConfigProto()
-config.gpu_options.per_process_gpu_memory_fraction = 0.9
+config.gpu_options.per_process_gpu_memory_fraction = 0.95
 #config.gpu_options.allow_growth=True   #不全部占满显存, 按需分配
 set_session(tf.Session(config=config))
 
@@ -44,6 +44,6 @@
 ms = ModelSpeech(datapath)
 
 #ms.LoadModel(modelpath + 'speech_model24_e_0_step_327500.model')
-ms.TrainModel(datapath, epoch = 50, batch_size = 4, save_step = 500)
+ms.TrainModel(datapath, epoch = 50, batch_size = 16, save_step = 500)