CSTR-Edinburgh / merlin

This is now the official location of the Merlin project.
http://www.cstr.ed.ac.uk/projects/merlin/
Apache License 2.0
1.31k stars 440 forks source link

Error in 05_train_duration_model.sh and 06_train_acoustic_model.sh #203

Open Rododendron82 opened 7 years ago

Rododendron82 commented 7 years ago

When I tried to build my own voice in the 5th and 6th steps I have this error: Traceback (most recent call last): File "/home/tarik/merlin-master/src/run_merlin.py", line 1158, in <module> main_function(cfg) File "/home/tarik/merlin-master/src/run_merlin.py", line 534, in main_function if cfg.VoiceConversion: AttributeError: 'configuration' object has no attribute 'VoiceConversion' It is related to master version.

rhajit1001 commented 7 years ago

When did you pull the project last time? IIRC they fixed it a few hours ago.

Rododendron82 commented 7 years ago

I have downloaded a few hours ago. Ok. I try a new version downloaded a few minuts ago. And there are new error: 2017-08-03 16:56:24,697 CRITICALListDataProvider: first list is empty 2017-08-03 16:56:24,697 CRITICAL main : train_DNN threw an exception Traceback (most recent call last): File "/home/tarik/merlin-master/src/run_merlin.py", line 1158, in <module> main_function(cfg) File "/home/tarik/merlin-master/src/run_merlin.py", line 817, in main_function cmp_mean_vector = cmp_mean_vector, cmp_std_vector = cmp_std_vector) File "/home/tarik/merlin-master/src/run_merlin.py", line 241, in train_DNN n_ins = n_ins, n_outs = n_outs, buffer_size = buffer_size, sequential = sequential_training, shuffle = False) File "/home/tarik/merlin-master/src/utils/providers.py", line 88, in __init__ assert len(x_file_list) > 0 AssertionError

ronanki commented 7 years ago

Did you change train, valid, test file number in conf/global_settings.cfg as per your data?

Rododendron82 commented 7 years ago

Did you change train, valid, test file number in conf/global_settings.cfg as per your

data?

No. Could you explain what values (train, valid, test) I have to enter?

ronanki commented 7 years ago

If you have a total of 1000 files, use 90% for training, and 5% each for valid, test. i.e., 900, 50, 50. If you have a total of 500 files, use 450, 25, 25 for train, valid and test respectively.

Rododendron82 commented 7 years ago

I have 500 wav and 500 txt files, but in label_state_align generated 250 files. So, I change global_settings.cfg: Train=200 Valid=15 Test=10 Script 05_train_duration_model.sh terminates without critical errors, but it have validation errors like this: 2017-08-04 10:35:32,585 DEBUG main.train_DNN: calculating validation loss 2017-08-04 10:35:32,823 INFO main.train_DNN: epoch 1, validation error 5.841296, train error 4.714705 time spent 6.14 After that, script 06_train_acoustic_model.sh returns critical error: IOError: [Errno 2] No such file or directory: '/home/tarik/merlin-master/egs/build_your_own_voice/s1/experiments/239/acoustic_model/data/mgc/p225_001.mgc' 4-5-steps-log.txt