A variable rate embedded-code speech waveform coder

Abstract

The design of variable rate coders for operation at 9.6 to 16 kbps that provide high speech quality and maintain robustness to environmental impairments, while retaining a low complexity, is an area of current research.An approach is here developed to achieve this objective, by exploiting a combination of Time Domain Harmonic Scaling algorithms and variable rate embedded-code ADPCM. The novel system, deeply examined and subjectively evaluated, emerges as a viable method for speech encoding, providing a quality equivalent to that of plain ADPCM at data rates of 24 to 48 kbps. INTRODUCTONVariable rate embedded-code digitizers are very attractive in a variety of speech processing applications such as packet switching, speech interpolation, voice-storage and message store-and-forward. The main feature which underlies the embedded-code structure, is that the rate change takes place on the digital channel and can be simply accomplished by deleting and inserting set of bits, without any elaborate code conversion. The actual data rate is determined by a control signal, according to the required network throughput, thus an effective management of link utilization, queue content and source activity is allowed. This paper deals with the design and performance analysis of an embedded-code Adaptive DPCM [1] combined with Time Domain Harmonic Scaling ITDHS) algorithms [2].Two main schemes are considered, with and without entropy coding, which operate in the range 9.6-16 kbps, with inputs bandlimited to 3.2 kHz and sampled at 6.4 kHz.Several important advantages emerge because of the waveform coding technique used, like high robustness to background noise and in tandem ing connection with other coders, good performance in presence of in-band non speech signals and, moreover, a low hardware complexity due to a computationally simple algorithm. System details are presented in Section 2. Section 3 discusses the experimental evaluation of the optimized codec. CODEC STRUCTUREA block diagram of the codec is shown in Fig. 1. Since ADPCM alone cannot provide acceptable quality at bit rates lower than about 16 kbps. , a rate-change algorithm is inserted as preprocessing end to halve the sampling rate of the input signal, so to enable the quantizer to use twice as many bits per sample.Even if a rigorous approach for frequency-scale modification of signals is based on the short-time Fourier analysis L3]. straightforward time-domain algorithms proposed in [21 seem to be more suitable for a simpler implementation without renouncing to speech quality. TDHS algorithms here exploited have been successfully combined with CVSD CH 1746•718210000 0212 $ 00.75 © 1982 IEEE 212[4], ARC [5], SBC and ATC systems [6], usually for scaling the speech spectrum of a factor of 2. TDHS AlgorithmsBasically, these algorithms change the speech rate by discarding or repeating short pieces of waveform which are at least equal in length to a pitch period. A refinement of a crude cut-and-splice method exploits the pitch information to perform a time-varying weightin...