Skip to content

Releases: gillesdegottex/pulsemodel

Version 2.3.0

05 Nov 14:50
Compare
Choose a tag to compare
  • Better and clearer management of situation when the window's length is bigger than the DFT's length.
  • More test coverage
  • Many little fixes about possible feature arguments

Version 2.2.4

12 Jun 09:51
Compare
Choose a tag to compare
  • WORLD is now the default spectral envelope estimator, without warning about article's resulting using STRAIGHT (though STRAIGHT will be used if straight_mcep is acessible).
  • Some renaming of arguments for consistency ... again
  • Force random seed to ensure reproducible tests
  • Add option for pre-resampling before analysis
  • Add option for pre-high-pass filtering before analysis
  • (move sigproc external to gitlab.com, remove REAPER's fork)

Version 2.1.0

06 Mar 21:36
Compare
Choose a tag to compare
  • Added frequency-warped cepstral compression

Version 2.0.5

01 Mar 23:32
Compare
Choose a tag to compare
  • Added frequency-warped compression
  • Synthesis: Added damping window on the end of the synthesis segment to avoid any possible signal cut during overlap-add (this only reduce risks of artifacts).
  • Synthesis: Now use 4 periods for each synthesis segments (reduces risks of artifacts; might be slightly slower)

Version 1.2.1

02 Oct 14:44
Compare
Choose a tag to compare

Now using WORLD vocoder amplitude spectral envelope
and REAPER f0 estimator is shipped with PML.
Support f0 binary file as input of analysis (thanks @ljuvela !).

Version 1.1.0

03 Sep 22:02
Compare
Choose a tag to compare

Now using WORLD vocoder amplitude spectral envelope.

Version 1.0.5

28 Aug 20:45
Compare
Choose a tag to compare
  • ATTENTION: Two options at synthesis have been renamed for consistency:
    --noisemaskfile/--nmfile
    --noisemask_nbbnds/--nm_nbbnds
  • ATTENTION: At synthesis, for consistency with the publications and with the behavior in the previous releases, nm_forcebinary has been replaced by !nm_cont
    I.e. The noise mask is always forced to binary values (with threshold 0.5),
    except if the option nm_cont is set to True (or the flag --nm_cont is used).
  • New option for attenuating the fist harmonic in silences since it is never randomized by PDD (see pp_atten1stharminsilences)
  • Apply a workaround to deal with waveforms whose F0 cannot be extracted (#11)
  • Can now output log f0 files during analysis
  • Can synthesis using mel-cepstral phase distortion deviation