Concatenation Issues

This technique successfully reverses the MFCC coding operation. The main weakness is that, because we threw away the harmonic information in the filter-hank step, we have to resort to a classical LP-style technique of using an impulse to drive the LP filter. Improvements to this have been made, with the motivation of generating a more-natural source, while still keeping a model system in which the parameters are largely statistically independent. For example, the technique of Yoshimiu a et al. [509] uses various excitation parameters that allow mixing of noise and impulse, and allow a degree of aperiodicity in the positions of the impulses. 

Having described a number of techniques for prosodic modifcation, we turn to the final issue in second-generation synthesis, that of how to join sections of waveform successfully, such that the joins cannot be heard so that the final speech sounds smoothly continuous and not obviously concatenated. 

Rgure 14.9 Phase problems can be caused by inconsistencies in epoch locations across join boundaries. 

While in second-generation synthesis signal processing is used mainly to modify pitch and timing, it can also be used in concatenation. If we are using a technique tiiat gives us some sort of spectral representation, such as residual-excited LP or sinusoidal modelling, then we can smooth or interpolate the spectral parameters at the join. This is possible only in models with a spectral representation, and is one of the reasons why residual-excited LP and sinusoidal models are chosen over PSOLA. 

Second-generation techniques are characterised by a hybrid approach of using data to determine the behaviour of the verbal or phonetic part of synthesis and an expUcit model plus signal processing to generate the correct prosody. 

---

We now turn to unit selection synthesis which is the dominant synthesis technique in text-to-speech today. Unit selection is the natural extension of second generation concatenative systems, and deals with the issues of how to manage large numbers of units, how to extend prosody beyond just FO and timing control, and how to alleviate the distortions caused by signal processing. 

---