-- QiongHu - 12 Dec 2012

For 12.12.12 Meeting

Figure

Attention: Here in the figure, using harmonic point to get the reconstructed spectrum is 110 points, 60 is the cepstrum order. If it hasn't changed in the figure, sorry for confusing.

• Unvoiced (all FFT points(1025 points) are selected for caculating the cepstrum coefficients (60 order))
  • Unwarped
    • Unwarped spectrum using SPTK: _sptk_unwarp_unvoice_nohar.png
    • Unwarped spectrum using Straight: _straight_unwarp_unvoice_nohar.png
  • Warped
    • warped spectrum using SPTK: _sptk_warp_unvoice_nohar.png
    • warped spectrum using Straight: _straight_warp_unvoice_nohar.png

• Voiced (all FFT points(1025 points) are selected for caculating the cepstrum coefficients (60 order))
  • Unwarped
    • Unwarped spectrum using SPTK: sptk_unwarp_voice_nohar.png
    • Unwarped spectrum using Straight: straight_unwarp_voice_nohar.png
  • Warped
    • warped spectrum using SPTK: sptk_warp_voice_nohar.png
    • warped spectrum using Straight: sptk_warp_voice_nohar.png

• Voiced (only harmonic points(110 points) are selected for caculating the cepstrum coefficients (60 order))( Just for comparing)
  • Unwarped
    • Unwarped spectrum using SPTK: sptk_unwarp_voice_har.png
    • Unwarped spectrum using Straight: sptk_unwarp_voice_har.png
  • Warped
    • warped spectrum using SPTK: sptk_warp_voice_har.png
    • warped spectrum using Straight: straight_warp_voice_har.png

Problem

1. Could the way which doesn't choose harmonic point be used for vocied spectrum caculation?
2. Warping or not doesn't greatly influence the result if using not harmonic point
3. Adding regulation or not doesn't greatly influence the result if using not harmonic point

Next

1. Produce voice from spectrum caculated from sinusodial way together with other parameter from STRAIGHT
2. caculate error between STRAIGHT spectrum, sinusodiul spectrum with the original one.
3. Caculating phase and adding jitter