[Giorgia M]

Hi Srikanth, yes, I believe this is what we did.

Step 1. Extract bottleneck features
Step 2. Append features to input and train new model. The error is lower than the system without bottlenecks, so it looks good.
Step 3. Trying to synthesise using weight matrices optimised during step 2.

The problem is that the weight matrices from step 2 are longer than the input that we use in step 3. The weight matrix has the length of the input + bottlenecks. But the input at synthesis time has the dimensionality of the input alone.

So the input now is a matrix (23,178) and the weight matrix is (274,1024). Cannot do the dot product.

[Giorgia M]

We have run the dnn_bottleneck.py file and stored the model.
Our understanding is that we need to use the hidden layer trained with the bottleneck model as input to the run_dnn.py. The bottleneck layer is a binary file in a directory: voices/…/LAYER_005_TANH_W.npy. Is it the file that we have to incorporate directly as the input or we have to convert it somehow to the lab format?

In relation to your previous answer, we have included an empty question file in the recipe-config. What does it mean “use the appended_input_dim as bottleneck dimension for the second DNN”? We have created a new variable in the config file, like this “appended_input_dim : 512”

Is this what you meant?

Thanks

[Giorgia M]

Thanks everyone.

Point 3 in Shrikanth’s post solved it. Bandmat installed on DICE machine!

[Giorgia M]

Hi Felipe,

The error thrown with lines 49, 53 and 54 uncommented is:

QUOTE
File “/afs/inf.ed.ac.uk/user/s15/s1566512/ossian_msc_2016_test/Ossian/scripts/processors/../../tools/dnn_tts/frontend/mlpg_fast.py”, line 53, in <module>
import bandmat as bm
ImportError: No module named bandmat
UNQUOTE

This looks like a vicious circle. I can’t escape bandmat! 😉

Thanks

[Giorgia M]

Still related to Bandmat:

I have completed the DNN training with 1 hour of speech (Italian, from Tundra). I am trying to synthesise now, but one of the files involved in synthesis is mlpg_fast.py which calls bandmat on various occasions. Specifically, the error I get is related to line 69:

Error QUOTE:

File “/afs/inf.ed.ac.uk/user/s15/s1566512/ossian_msc_2016_test/Ossian/scripts/processors/../../tools/dnn_tts/frontend/mlpg_fast.py”, line 69, in build_win_mats
win_mat = bm.band_c_bm(u, l, win_coeffs).T
NameError: global name ‘bm’ is not defined

Error UNQUOTE

If I replace

win_mat = bm.band_c_bm(u, l, win_coeffs).T

with

win_mat = pyximport.band_c_bm(u, l, win_coeffs).T

I get this error:

QUOTE

File “/afs/inf.ed.ac.uk/user/s15/s1566512/ossian_msc_2016_test/Ossian/scripts/processors/../../tools/dnn_tts/frontend/mlpg_fast.py”, line 69, in build_win_mats
win_mat = pyximport.band_c_bm(u, l, win_coeffs).T
AttributeError: ‘module’ object has no attribute ‘band_c_bm’

UNQUOTE

How can I go ahead?

Thanks

[Giorgia M]

Solved by:

1. uncommenting line 49 of the file mlpg_fast.py
2. commenting lines 53 and 54 of the same file
3. installing the full cython package (it wouldn’t let install only pyximport): pip install –user cython

Thanks

[Giorgia M]

Hi, I am using the iMac now (in the Psychology basement), and I get the same error. See the error thrown at DNN training.

QUOTE

ppls-atlab-010:Ossian s1566512$ python ./scripts/train.py -s rss_toy_demo -l rm msc_demo_02
— Gather corpus
— Train voice
/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/train//rm/speakers/rss_toy_demo/msc_demo_02
/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/voices//rm/rss_toy_demo/msc_demo_02
Load voice from /mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/recipes/msc_demo_02.cfg …
*** voice loading processor tokeniser ***
BasicTokenisers.RegexTokeniser —
*** voice loading processor token_classifier ***
BasicTokenisers.RegexClassifier —
*** voice loading processor token_safetexter ***
BasicTokenisers.SafeTextMaker —
*** voice loading processor segment_adder ***
BasicTokenisers.RegexTokeniser —
*** voice loading processor letter_safetexter ***
BasicTokenisers.SafeTextMaker —
*** voice loading processor end_silence_adder ***
SimpleChildAdder.SimpleChildAdder —
*** voice loading processor letter_vsm ***
VSMTagger.VSMTagger —
*** voice loading processor word_vsm ***
VSMTagger.VSMTagger —
*** voice loading processor punc_adder ***
SimpleChildAdder.SimpleChildAdder —
*** voice loading processor space_adder ***
SimpleChildAdder.SimpleChildAdder —
*** voice loading processor align_label_dumper ***
FeatureDumper.FeatureDumper —
*** voice loading processor speech_feature_extractor ***
FeatureExtractor.WorldExtractor —
*** voice loading processor aligner ***
Aligner.StateAligner —
*** voice loading processor pause_predictor ***
SKLProcessors.SKLDecisionTree —
*** voice loading processor phrase_adder ***
PhraseMaker.PhraseMaker —
*** voice loading processor dur_data_maker ***
FeatureDumper.FeatureDumper —
*** voice loading processor dur_label_maker ***
FeatureDumper.FeatureDumper —
*** voice loading processor duration_predictor ***
NN.NNDurationPredictor —
Traceback (most recent call last):
File “./scripts/train.py”, line 147, in <module>
main_work()
File “./scripts/train.py”, line 82, in main_work
train(opts, dirs)
File “./scripts/train.py”, line 118, in train
dirs, clear_old_data=opts.clear, max_cores=opts.max_cores)
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/main/Voice.py”, line 104, in __init__
self.config[self.run_mode].walk(self._load_item, call_on_sections=False)
File “/Users/s1566512/Library/Python/2.7/lib/python/site-packages/configobj.py”, line 894, in walk
val = function(self, entry, **keywargs)
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/main/Voice.py”, line 117, in _load_item
self.config[stage].walk(self._load_item, call_on_sections=False)
File “/Users/s1566512/Library/Python/2.7/lib/python/site-packages/configobj.py”, line 894, in walk
val = function(self, entry, **keywargs)
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/main/Voice.py”, line 123, in _load_item
self.load_processor(proc_name)
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/main/Voice.py”, line 167, in load_processor
ClassName = dynamic_load_object(class_string) ## , model=model)
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/main/Voice.py”, line 386, in dynamic_load_object
m = __import__(module )
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/processors/NN.py”, line 35, in <module>
from frontend.mlpg_fast import MLParameterGenerationFast
File “/mnt/courses.homes/s1566512/Desktop/ossian_msc_2016_test/Ossian/scripts/processors/../../tools/dnn_tts/frontend/mlpg_fast.py”, line 53, in <module>
import bandmat as bm
ImportError: No module named bandmat

UNQUOTE

Thanks

[Giorgia M]

I managed to do this. Seems like training is working now.

Thanks,
Giorgia

[Giorgia M]

Thanks, I have troubles with the HMM training. Not in DNN yet.

Following the instructions, I have created a shortlink called “OSSIAN” to the Ossian path:

/afs/inf.ed.ac.uk/group/cstr/projects/simple4all_2/oliver/temp/ossian_msc_2016_test/

What do you mean by “Change the Ossian path in the below files and configure it to your own local path”?

Thanks

[Author]

Posts