Introducing compact: An oscillator-based approach to toll-quality speech coding at low bit rates

• Published in: 2010 - MILCOM 2010 MILITARY COMMUNICATIONS CONFERENCE pp. 293 - 297
• Main Authors: Yen, A Y, Gorodnitsky, I
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.10.2010
• Subjects: Audio oscillators; Bit rate; Delay; Mathematical model; Oscillators; Signal to noise ratio; Speech; speech codecs; Speech coding; speech processing; speech synthesis
• ISBN: 9781424481781; 1424481783
• ISSN: 2155-7578; 2155-7586
• DOI: 10.1109/MILCOM.2010.5680310

In this paper, we introduce an improved oscillator model we term the Complete Oscillator Model (COM). A significant advantage of the COM over classical oscillators such as the Self Excited Vocoder is that it is not restricted to modeling only certain larger-scale patterns in the source sequence. Here, we develop a speech coding system based on the proposed COM. In this system, the COM is used in combination with a linear predictor, the Pulsed Autoregressive CompensaTor (PACT), to develop a novel, oscillator-based approach to toll-quality speech coding at low bit rates. Unlike the linear prediction-based models utilized in modern speech coders, oscillators do not depend on an estimate of the residual error to regenerate the signal. As such, the residual is encoded only for select frames, providing a potential improvement in coding efficiency. An implementation of the hybrid COM/PACT system, which we call COMPACT, is described and is shown to provide both perceptual quality and bit rate that are competitive with mature standards such as G.729 and AMR. The given implementation is demonstrated to produce toll-quality speech, as measured by PESQ-MOS, at 9.77 kbps. Future tuning of this implementation is expected to improve performance to where it could exceed the current state of the art.