Hi, I'm trying to extract features from my own dataset but I find that you use interpolate.py during alignment, which isn't present in the repository or elsewhere. Could you please add the file and/or describe what it does?

I tried the solution provided by lf0_lstm.py and so. When I tried to modify the parameters in tranning, a script in /data/training/compute_dtw.sh made me confused.

` ZASKA="Zaska -P $PRM_NAME $PRM_OPT"

# Compute mfcc $DIR_REF/${FILENAME}.wav $DIR_TST/${FILENAME}.wav => mfcc/$DIR_REF/${FILENAME}.prm mfcc/$DIR_TST/${FILENAME}.prm
${ZASKA} -t RAW -x wav=msw -n . -p mfcc -F ${DIR_REF}/${FILENAME}_sil ${DIR_TST}/${FILENAME}_sil

# Align: mfcc/${DIR_REF}/${FILENAME}.prm, mfcc/${DIR_TST/${FILENAME}.prm => dtw/${DIR_REF}-${DIR_TST}/${FILENAME}.dtw
b=2
dtw -b -${b} -t mfcc/${DIR_TST} -r mfcc/${DIR_REF} -a ${DIR_DTW}/beam${b} -w -B -f -F ${FILENAME}_sil`

Running the script is diffcult, as the command "Zaska" is not exist in any package I found and the "dtw" command doesnt have the parameter of "-b" . How could I sovle it?

By the way, I want to run this script because I wanted to add more parameters on training, I modified "tfglib" and tried to build the /data/train_datatable.h5 again.

It resulting in very few harmonic elements, may need to use high-order of feature extraction and adjust the network to fit more high-order implied feature.(It seems the default training also resulting over-fitting.)
In addition, the result of converted result sounds dull and low, lacks a sense of penetration, may due to the lack of high-order features harmonic elements.

MR albert, I love your work so much.
but it doesn't have a friendly user guide to use.
and....it may have some bugs in these codes?
these day I am trying to run these codes but I spent some time on how to use it. May I share my note and could you please spent a little time checking on it so that we could build it more friendly?
Here is my note, it seems someting wrong....（On step 6 when i using seq2seq） butI dont know where is the problem.

This project contains 3 solutions for voice conversion.
A MCEP-GMM based solution based on SPTK tools（Bsaeline solution of VCC2016: http://vc-challenge.org/summary.html）
A DNN-LSTM-GRU converting MVF-logf0-Mel-cepstrum
A seq2seq based MVF-logf0-Mel-cepstrum feature extraction conveting solution.

To run MCEP-GMM based solution based on SPTK tools:
edit sptk_vc.sh TRAIN_FILENAME to any files you need to convert
run sptk_vc.sh
the output wav is in data/training/gmm_vc

To run DNN-LSTM-GRU converting:
run lf0_lstm.py mvf_dnn.py mcp_gru.py to train different models to convert different features
[optional]run lf0_post_training.py mcp_post_training.py mvf_post_training.py mvf_plot_curves.py to verify model
run decode_aho.sh to merge the feature to wav

To run seq2seq:
0.[optional]you can get all the training file in ,put them in data/training/
1.apt-get intsall sox ,pip install tensorflow and so on
2.cp do_columns.pl to /usr/local/bin
3.get tfglib (https://github.com/albertaparicio/tfglib),edit seq2seq_datatable.py(maybe bug in para: nb_classes) ,and install
4.source install ahocoder, and add the file $ into your path
5.edit data/test/speakers.list (add more speakers if step 0 was procceed?)
6.run /data/train/seq2seq_align_training.sh and /data/test/seq2seq_align_test.sh 
7.run seq2seq.py
(There are some questions....that some file(like file 200007)wasn't extract .lf0.dat file and may throw an error)
8.run seq2seq_decode_prediction.py

When I try to run decode_aho.sh, I receive this error. Is it an error of the data format?

Convert parameters to float data
usage: x2x [-f base] [-t base] [-l | -u] value
x2x: error: unrecognized arguments: data/test/predicted/SF1-TF1/200005.lf0.dat

Thank you

Thank you for your great work.
And，
I run data/training/seq2seq_align_training.sh . And got an error showing that it needs a interpolate.py which is not in my system. Could u show me how to get the interpolate.py？

Here is the code in data/training/seq2seq_align_training.sh：
# Interpolate lfo and vf data
python $(which interpolate.py)
--f0_file ${DIR_VOC}/${DIR_SPK}/${FILENAME}.lf0.dat
--vf_file ${DIR_VOC}/${DIR_SPK}/${FILENAME}.vf.dat
--no-uv

run with error of：
Processing SM2/200040
Unknown option: --
usage: python [option] ... [-c cmd | -m mod | file | -] [arg] ...
Try `python -h' for more information.

I search for interpolate.py and found scipy has such a file ,but without parameter of f0_file vf_file and no-uv_
how to get the code work?THX

Hi,
According to mcp_gru.py, the input batch has the size of (batch_size, tsteps, data_dim) with batch_size = 1 and tsteps=50. So I guess after DTW, we have to chop the source and targets files into portions with the size of (tsteps, data_dim). Is it right?
Are those portions (in the same file) overlap?
Thanks.

Dear Mr albert ,I am a student from China .
I am interested in your work very much. But I can't download the file you shared for some reason said "the link is failure". So Can you share a new link ? I wish you can offer a friendly user guide to use too.
Thank you very much !
It also would be great that if anyone who have download the files successfully before can share them to me.

Hello,

I'm sorry to interrupt you, I'm interested in your mcep-gmm method to do the voice conversion and I have executed your program with my computer, but I find when I established a 10 Gaussian components model and a 50 Gaussian components model respectively, the results of converting wav file 20007.wav keep always unchangeable, so I want to ask how many Gaussian components and iteration you have used ?
Also, the number of Gaussian component in the command vc and the number of component in the command gmm should they keep the same number ?

Thank you for your attention
J.SHI