bug during training with new voice data

Hi, I replaced the slt_demo data with a new voice data, following the instruction given here. All the scripts for data prepreration went smoothly. However, during the Step2 : training duration model, I got the following error info. Any idea or advice? Thanks a lot!

2017-01-02 00:34:49,253 DEBUG acoustic_comp: wrote 45 frames of features 2017-01-02 00:34:49,869 INFO main : normalising acoustic (output) features using method MVN /home/merlin/src/frontend/mean_variance_norm.py:69: FutureWarning: comparison to None will result in an elementwise object comparison in the future. if self.mean_vector == None: /home/merlin/src/frontend/mean_variance_norm.py:71: FutureWarning: comparison to None will result in an elementwise object comparison in the future. if self.std_vector == None: 2017-01-02 00:34:49,940 INFO main : saved MVN vectors to /home/merlin/egs/slt_arctic/s1/experiments/slt_arctic_demo/duration_model/data/norm_info_dur_5_MVN.dat 2017-01-02 00:34:49,941 INFO main : saved dur variance vector to /home//merlin/egs/slt_arctic/s1/experiments/slt_arctic_demo/duration_model/data/var/dur_5 2017-01-02 00:34:49,958 DEBUG labels : HTS-derived input feature dimension is 416 + 0 = 416 2017-01-02 00:34:49,958 INFO main : label dimension is 416 2017-01-02 00:34:49,959 INFO main : training DNN 2017-01-02 00:34:49,959 DEBUG main.train_DNN: Starting train_DNN 2017-01-02 00:34:49,959 DEBUG main.train_DNN: Creating training data provider 2017-01-02 00:34:49,960 DEBUG main.train_DNN: Creating validation data provider 2017-01-02 00:34:49,960 CRITICALListDataProvider: first list is empty 2017-01-02 00:34:49,960 CRITICAL main : train_DNN threw an exception Traceback (most recent call last): File “/home//merlin/src/run_merlin.py”, line 1175, in <module> main_function(cfg) File “/home/merlin/src/run_merlin.py”, line 838, in main_function cmp_mean_vector = cmp_mean_vector, cmp_std_vector = cmp_std_vector) File “/home//merlin/src/run_merlin.py”, line 209, in train_DNN n_ins = n_ins, n_outs = n_outs, buffer_size = buffer_size, sequential = sequential_training, shuffle = False) File “/home/merlin/src/utils/providers.py”, line 88, in init assert len(x_file_list) > 0 AssertionError

[zhizhengwu]

(train_file_number + valid_file_number + test_file_number) should be smaller than the total number of files.

On 4 January 2017 at 14:01, Jun ZHANG [e-mail] wrote:

yes. I actually followed the demo example, in the duration_slt_arctic_demo.conf, I specified the following " [Data] train_file_number: 417 valid_file_number: 5 test_file_number: 5"

note the total number of files in the new voice dataset is 417. Very confused how the validation data became “empty”? Any further advice for the bug ? I highly appreciate the effort 😃

— You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub https://github.com/CSTR-Edinburgh/merlin/issues/69#issuecomment-270499685, or mute the thread https://github.com/notifications/unsubscribe-auth/AFRxxpl3kkpGSE4C5e7ZrQU6xLxGliGQks5rPBbKgaJpZM4LZXF0 .

[ikhfa]

change the train_file_number to 407