A 9.6-kb/s DSP Speech Coder

The subject of digital speech encoding and bit-rate compression has been one of considerable interest in recent years. Attention has focused strongly on bit rates in the range of 9.6 kb/s for applications where good "communications quality" is required and where robustness across a broad range of background noise conditions and speaker variations is necessary.12 This bit rate appears, at present, to be about the lowest practical rate at which this standard of quality and robustness can be reliably achieved. Below 9.6 kb/s presently known techniques have a noticeable synthetic quality and are considerably more fragile to differences in speakers and background conditions.1 Several encoding methods for achieving "communications quality" at 9.6 kb/s have been proposed and studied. Most of these methods involve a considerable amount of signal processing to meet these goals and are thus referred to as "high-complexity" algorithms. Their imple2263 mentation in real time typically requires the use of specially designed high-speed digital hardware or array-processing digital computers.3,4 In recent studies,2 we found that a combined technique of sub-band coding (SBC) and time-domain harmonic scaling (TDHS) leads to an encoding approach whose quality is comparable to, or better than, some of the previously studied "high-complexity" algorithms. The interesting aspect of this approach is that it is a combination of two relatively "low-complexity" algorithms that are amenable to real-time implementation using currently available technology.