-- QiongHu - 19 Dec 2012

Figures&Waves

  1. (Up)Reconstructed sinusodial spectrum using SPTK to caculate the errors.
  2. (Down) Reconstructed sinusodial spectrum using STRAIGHT to caculate the errors.
Open the figure here:Figure.png
  • red line: the reconstructed spectrum
  • blue line: the original spectrum from Straight
  • yellow line: the original spectrum from STPK
  • green line: the spectrum after using the mel-cepstrum to spectrum using STRAIGHT

speech  
original speech listen
straight spectrum--> mcep --> mcep2sp --> STRAIGHT synthesis listen
FFT Spectra -- > Non-frequency-warped Discrete cepstrum --> Reconstructed spectrum --> mcep --> mcep2sp --> STRAIGHT synthesis listen
straight spectrum-- > Non-frequency-warped Discrete cepstrum --> Reconstructed spectrum --> mcep --> mcep2sp --> STRAIGHT synthesis

listen

straight spectrum --> STRAIGHT synthesis listen
FFT Spectra -- > Non-frequency-warped Discrete cepstrum --> Reconstructed spectrum --> STRAIGHT synthesis listen
straight spectrum-- > Non-frequency-warped Discrete cepstrum --> Reconstructed spectrum --> STRAIGHT synthesis listen

Notice:

  • Here the way to synthesize speech is to use all FFT point(not harmonic) to caculate the error function in the last two ways.
  • vocoder is the straight one, and phase and aperiodic features are using Sraight ones.

Analysis&Problem:

  • Wave using spectrum from SPTK sounds worse
  • If all points are using, Warped or unwarped caculated spectrums look similar.
  • The difference STRAIGHT mel-cepstrum to smoothed spectrum with sinusodial warped spectrum???
  • normal cepctrum? (STPK mcep)
  • Not using harmonics point: as accuracy of peak selection greatly affects the result.
  • The reason to use harmonic point instead of all the points. *
Topic attachments
I Attachment Action Size Date Who Comment
wavwav cmu_us_arctic_slt_a0011.wav manage 286.0 K 19 Dec 2012 - 03:02 QiongHu  
elseori cmu_us_arctic_slt_a0011.wav.ori manage 285.5 K 19 Dec 2012 - 03:01 QiongHu  
elseunwarp cmu_us_arctic_slt_a0011.wav.ori.unwarp manage 285.5 K 19 Dec 2012 - 04:33 QiongHu  
elsesptk cmu_us_arctic_slt_a0011.wav.sptk manage 285.5 K 19 Dec 2012 - 03:01 QiongHu  
elseunwarp cmu_us_arctic_slt_a0011.wav.sptk.unwarp manage 285.5 K 19 Dec 2012 - 04:33 QiongHu  
elsestraight cmu_us_arctic_slt_a0011.wav.straight manage 285.5 K 19 Dec 2012 - 03:46 QiongHu  
elseunwarp cmu_us_arctic_slt_a0011.wav.straight.unwarp manage 285.5 K 19 Dec 2012 - 04:34 QiongHu  
pngpng reconstruted.png manage 25.0 K 19 Dec 2012 - 02:45 QiongHu  
Topic revision: r4 - 14 Feb 2013 - 09:26:48 - QiongHu
 
This site is powered by the TWiki collaboration platformCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding TWiki? Send feedback
This Wiki uses Cookies