search

bajibabu / GlottGAN

This repository contains the files used for our Interspeech 2017 paper.
16 stars 6 forks source link

How to prepare data for training #1

Open tuanad121 opened 7 years ago

tuanad121 commented 7 years ago

Is there any instruction about prepare the training data? Thanks for spending your time on my issue. I appreciate it.

ljuvela commented 7 years ago

I uploaded some pre-packaged data to https://drive.google.com/open?id=0B5M0_b2kZdj6MVFCU2JpX1NON0U

The tools we used for data extraction are included in our soon-to-be-released GlottDNN vocoder. I'll make sure to update here once the code is out.

It's also feasible (but a lot of work) to build a data extractor yourself. You would first need 1) A glottal inverse filtering method 2) Pitch marks (glottal closure instants) Some openly available tools for that can be found here: https://github.com/covarep/covarep/tree/master/glottalsource

Second, you would need to window the target signals for neural net output. See this paper (especially Fig. 2) on how to do that. https://www.researchgate.net/publication/298710465_High-pitched_excitation_generation_for_glottal_vocoding_in_statistical_parametric_speech_synthesis_using_a_deep_neural_network

Hope this helps!

tuanad121 commented 7 years ago

I uploaded some pre-packaged data to

Great, I appreciate it.

The tools we used for data extraction are included in our soon-to-be-released GlottDNN vocoder. I'll make sure to update here once the code is out.

It's nice. Hopefully, it has python (or matlab) version ^^

I will follow your instructions. Thanks for your help.

licktion commented 6 years ago

@ljuvela Just curious about if your GlottDNN vocoder has been released or not? Thanks!