TABLE OF CONTENTS This chapter presented a survey of the main techniques in speech communications primarily related to the quantisation of the speech signal, and the trade-offs that this involves in terms of quality and computational complexity. Both time-domain waveform quantisation and parameterisation were discussed, before a presentation of pitch modelling and extraction techniques. Finally, the important and pragmatic audio engineering subject of analysis-by-synthesis was explored.
Next, the major remaining methods of audio analysis (that have not already been covered) will be presented in Chapter 6, before Chapter 7 rounds off with advanced topics that make use of many of the techniques discussed in the current and previous chapters.
There is an entire spectrum of speech coding and compression systems, and then another entire spectrum designed to handle music and general audio. The tendency is for the higher rate algorithms to have higher perceived quality but there are also significant differentiating factors in the computational complexity, and in the underlying techniques used: the class of CELP-based systems contains many weird and wonderful performance-enhancing ideas in addition to the basic codebook excited linear prediction structure.
| Name | Description |
|---|---|
| G.723.1 | The basic ITU CELP standard at 5.3 or 6.3 kbits/s |
| G.728 | Low delay CELP, 16 kbits/s sampled at 8 kHz |
| G.729 | Algebraic CELP 8 kbits/s standard |
| TETRA | CELP 4.8 kbits/s |
| FS1015 | LPC-based tenth-order system at 2.4 kbits/s |
| FS1016 |
