VoiceConversion

VoiceConverision.jl is a statistical voice conversion library for Julia.

Please note that this package is still under developement. Both bug reports and feature requests are welcome.

Features

Parameter conversion

• Frame-by-frame parameter conversion using joint Gaussian Mixture Models (GMMs) of source and target feature space src/gmmmap.jl
• Trajectory parameter conversion based on maximum likelihood criterion w/o considering Gloval Variance (GV) [Toda 2007] src/trajectory_gmmmap.jl

Waveform modification

• WORLD-based vocoding
• Direct waveform modification using Mel log Spectrum Approximation (MLSA) digital filter based on spectrum differencial [Kobayashi 2014]

Supported Platforms

• Linux
• Mac OS X
• Windows

Installation

Run the following commands on your julia interactive settion (REPL):

julia> Pkg.clone("https://github.com/r9y9/VoiceConversion.jl")

This should resolve package dependencies property.

Additional requirements for CLI programs

See bin/REQUIRE.

Demonstration using CMU Arctic

Please check examples/cmu_arctic/cmu_arctic_demo.jl.

References

• [Toda 2007] T. Toda, A. W. Black, and K. Tokuda, “Voice conversion based on maximum likelihood estimation of spectral parameter trajectory,” IEEE Trans. Audio, Speech, Lang. Process, vol. 15, no. 8, pp. 2222–2235, Nov. 2007.
• [Kobayashi 2014] Kobayashi, Kazuhiro, et al. "Statistical Singing Voice Conversion with Direct Waveform Modification based on the Spectrum Differential." Fifteenth Annual Conference of the International Speech Communication Association. 2014.