The Channel Vocoder

One way to approximately estimate energy without actually computing Equation 7.1 is to sum the magnitude of the samples  

When used to encode and process speech, the channel vocoder explicitly makes an assumption that the signal being modeled is a single human voice. The source analysis block extracts some parameters about the fine spectral detail, such as whether the sound is pitched (vowel) or noisy (consonant or 

---

The phase vocoder has been useful in a number of applications4. In time-scale modification, for example, the goal is to maintain the perceptual quality of the original signal while changing its apparent rate of articulation . In performing time-scale modification with the phase vocoder, the instantaneous frequency and amplitude of each channel are interpolated or decimated to a new time scale . In one scenario, the phase of each filter output in Equation (9.9) is first unwrapped, and the channel amplitude and unwrapped phase are then time scaled. With time-scale modification by a factor p, the modified filter output is given by... 

Since each harmonic is a multiple of the time-varying fundamental, higher harmonics vibrato have a larger bandwidth than lower harmonics. With rapid pitch vibrato, the temporal resolution required for frequency estimation increases with harmonic number. One approach to improve resolution time-warps the waveform inversely to pitch to remove vibrato [Ramalho, 1994]. This approach may also be useful in reducing channel cross-talk within the phase vocoder. 

Figure 7.6 shows a block diagram of a channel vocoder. The detected envelopes serve as control signals for a bank of bandpass synthesis filters (identical to the analysis filters used to extract the subband envelopes). The synthesis filters have gain inputs that are fed by the control signals. 

One ofthe most widely used techniques is the STRAIGHT set of analysis tools [ ].This system is operates as high quality speech analysis-modification-synthesis method implemented as a channel vocoder and has separate components for instantaneous-ffequency-based FO extraction and pitch-adaptive spectral smoothing. STRAIGHT attempts to obtain a more accurate spectral estimation and a use more sophisticated soiuce model than simple impulses. A comparison of STRAIGHT and standard cepstral analysis is showen in Figure 15.17. 

Based on the source-filter modeling, the spectral envelope, which is the smooth approximation of the magnitude spectrum shape, can be estimated [1], Many methods such as cepstrum, linear predictive coefficient, LPC and channel vocoder method can be used to estimate the spectral envelope [1 10], In this paper, the cepstrum method is used to estimate the spectral envelope. 

The size of the window defines the number of input samples to be analysed at a time. The larger the window, the greater the number of channels, but the lower the time resolution, and vice versa. This should be set large enough to capture four periods of the lowest frequency of interest. The sampling rate divided by the number of channels should be less than the lowest pitch in the input sound. This may be set in terms of the lowest frequency of interest or in terms of the number of channels required. Note that the term channel is the Phase Vocoder jargon for what we referred earlier to as harmonic detectors (Figure 3.11). 

Figure 3.11 The Phase Vocoder uses channels to act as harmonic detectors . The greater the number of channels, the better the frequency resolution. Note that (b) does not distinguish between the two uppermost partials that are perfectly distinct In (a)...

---