Academia.eduAcademia.edu

Figure 5 – uploaded by Petros Maragos

See full PDFdownloadDownload figure
In this case, the DFT coefficient makes a 1 jump unwrapped phase at the instant n by taking the previous phase as the current phase: v(n,k) =v(n —1,k) to avoid the chaotic phase jump. This constitutes the third modification introduced into the Portnoff algorithm as previously stated in section 2 above.

Figure 5 In this case, the DFT coefficient makes a 1 jump unwrapped phase at the instant n by taking the previous phase as the current phase: v(n,k) =v(n —1,k) to avoid the chaotic phase jump. This constitutes the third modification introduced into the Portnoff algorithm as previously stated in section 2 above.

Related Figures (5)

3. In an algorithm of this type based on phase vocoder techniques, very great care must taken in performing the phase unwrapping. In order to calculate the term v(n, w), we implemented a phase unwrapping procedure which expresses v(n,W) only in terms of v(n-—Il,w), y(n -1,w) and y(n,w) where y(n,W) is the instantaneous phase of the DFT coefficient at the instant n and frequency wW . The next section will give the explicit calculation for v(n, W). Since D and I must be small, the set of the values of B is quite limited in size. The action of the zero-padding operator 1:I which inserts I-1 zeroes between two DFT coefficients belonging to the same frequency channel, followed by that of the interpolating filter f,,(m) and the decimating operator D:1, results in the spectral level time- scale modification of the signal. The phase is multiplied by 1/ B —1 instead of 1/ B in orderto avoid that the phase multiplication affects the phase modulation term. followed by that of the interpolating filter f,,(m) and the
3. In an algorithm of this type based on phase vocoder techniques, very great care must taken in performing the phase unwrapping. In order to calculate the term v(n, w), we implemented a phase unwrapping procedure which expresses v(n,W) only in terms of v(n-—Il,w), y(n -1,w) and y(n,w) where y(n,W) is the instantaneous phase of the DFT coefficient at the instant n and frequency wW . The next section will give the explicit calculation for v(n, W). Since D and I must be small, the set of the values of B is quite limited in size. The action of the zero-padding operator 1:I which inserts I-1 zeroes between two DFT coefficients belonging to the same frequency channel, followed by that of the interpolating filter f,,(m) and the decimating operator D:1, results in the spectral level time- scale modification of the signal. The phase is multiplied by 1/ B —1 instead of 1/ B in orderto avoid that the phase multiplication affects the phase modulation term. followed by that of the interpolating filter f,,(m) and the
The interest of Lagrange interpolating filters for this type of application resides in the facts that they are zero phase and preserve the support from which the interpolated points are calculated.
The interest of Lagrange interpolating filters for this type of application resides in the facts that they are zero phase and preserve the support from which the interpolated points are calculated.
A chaotic phase jump was observed in the region close to the origin. The phase unwrapping procedure must take this into account
A chaotic phase jump was observed in the region close to the origin. The phase unwrapping procedure must take this into account
In this case, the DFT coefficient makes a 2M jump
In this case, the DFT coefficient makes a 2M jump
Related topics:
Cognitive ScienceComputer ScienceSpeech SynthesisSpeech CommunicationSpeech ProcessingSpeech analysisFilter BankFrequency ModulationAmplitude ModulationPitch Tracking
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved