Epoch detection

To perform closed phase analysis properly, we need to determine when the glottis shuts these points are called the instants of glottal closure and can be found automatically by the epoch detection algorithms described below in Section 12.7.1. A further benefit to performing closed phase analysis is that during these intervals, the closed glottis obeys the model assumptions of... 

Epoch detection finding the instant of glottal closure... 

The above technique bears some similarities to the TD-PSOLA technique in that it uses a pitch-synchronous analysis to isolate individual pitch periods, after which modification and resynihesis is performed. In fact in a technique called linear prediction pitch synchronous overlap and add or LP-PSOLA, we can use the PSOLA more or less directly on the residual rather than the waveform. As above, epoch detection is used to find the epochs. The residual is then separated into a number of symmetrical frames centred on the epoch. Pitch modification is performed by moving the residual frames closer or further away, and duration modification is performed by duplication or elimination of frames, in just the same way as in TD-PSOLA. The only difference is that these operations are performed on the residual which is then fed into the LP filter to produce speech. This technique differs only from the Hunt technique in the shape of the frames. Both techniques uses window functions with their highest point at the epoch in Hunt s technique the windows are asymmetrical with the idea that they are capturing a single impulse, in LP-PSOLA the windows are symmetrical. In listening tests, the two techniques produced virtually identical quality speech. 

Dutoit [148] investigates the issue of epoch accuracy and concludes that the problem is complicated, partly due to a lack of formal definition as to where exactly the epoch should lie (the instant of glottal closure is only one possibility). Since then, more accurate epoch detection algorithms have been developed (see Section 12.7.2) but there are cases (for example with poorer quality recordings) where perfect automatic epoch detection may not be possible. 

MBROLA is a PSOLA like technique which uses sinusoidal modelling to decompose each frame and from this resynthesise the database at a constant pitch and phase, thus alleviating many problems in inaccurate epoch detection. 

We have just seen that closed-phase LP requires that we analyse each pitch period separately. This type of speech analysis is called pitch-synchronous analysis and can be performed only if we are in fact able to find and isolate individual periods of speech. We do this by means of a pitch-marking or epoch-detection algorithm (EDA). 

The above technique bears some similarities to the TD-PSOLA technique in that it uses a pitch-synchronous analysis to isolate individual pitch periods, after which modification and resynthesis are performed. In fact, in a technique called linear-prediction pitch-synchronous overlap and add or LP-PSOLA, we can use PSOLA more or less directly on the residual rather than on the waveform. As above, epoch detection is used to find the epochs. The residual is then separated into a number of symmetrical frames centred on... 

One significant problem in many types of synthesis is that there may be phase mismatches between units. An example of this is shown in Figure 14.9, where we have a situation in which botii units have accurate epochs, resulting in correct pitch-synchronous analysis, but the point of reference for the epochs is different for each unit. When concatenated, this will result in a single irregular period at the join, which is usually detectable. Many proposals have been made as to how to fix this problem [149]. Obviously, we can try to ensure consistent epoch detection in the first place. Good epoch detectors are now more common, but this was not always possible in the past. Other solutions include a... 

---