WavGenSR

Waveform generator based on signal reshaping for statistical parametric speech synthesis.

This software is based on the work presented at Interspeech 2016:

F. Espic, C. Valentini-Botinhao, Z. Wu, and S. King, “Waveform generation based on signal reshaping for statistical parametric speech synthesis,” in Proc. Interspeech, San Francisco, CA, USA, September, 2016.

More information at http://felipeespic.com/wavgensr/

@author: Felipe Espic

I. Description

This software generates speech waveforms from the speech features:

1. Spectral envelope
2. F0 contour

, which may be extracted by a vocoder (e.g., STRAIGHT, WORLD) or generated by an acoustic model (e.g., Merlin, HTS).

Key points:

• No need of aperiodicities.
• No minimum phase assumtion.
• Natural phase spectrum plus periodic and aperiodic components given "for free" by natural speech recordings.

II. License

See the LICENCE file for details.

III. Prerequisites

• OS: Linux (not tested in other OS's yet)
• MATLAB (Any modern version)

IV. Install

Just download and compile SPTK by:

cd tools
./download_and_compile_sptk.sh

This will compile SPTK automatically for you.

V. Usage

Just go to /demos, read the instructions in the WavGenSR_demo.m script, and run it using MATLAB.

You will lern very quickly how to use the main function WavGenSR by this simple demo.

Then, if wanted you can modify WavGenSR_demo.m to suit your needs.