【摘要】 矢量量化因其很高的压缩率被广泛应用在语音压缩编码系统中,抗误码能力差是影响其在高误码信道下性能的主要原因。信道优化矢量量化(channel optimized vector quantization,COVQ)能有效降低有误码情况下的量化误差,但在无误码时的量化性能却不如传统的矢量量化。为了提高矢量量化对信道误码的鲁棒性,同时又不影响无误码时的量化性能,该文提出了一种COVQ与非等重保护(unequal error protection,UEP)相结合的信源信道联合编码方案。该方案利用了COVQ在抗误码方面的优势,同时与UEP有效地结合,不仅能提高COVQ在无误码时的量化性能,而且还能进一步提升抗误码能力。在实际语音编码系统中的仿真测试表明:该方案不仅在无误码时降低了因COVQ导致的量化失真约50%,而且在误码率较高时最后合成语音的平均意见得分(mean opinion score,MOS)比传统的信源信道方案有最大0.58的提高。更多还原

【Abstract】 Vector quantization is widely used in speech coding systems because of its high compression ability.However,its poor anti-error capability greatly affects its performance in noisy channels.Channel optimized vector quantization(COVQ) can reduce quantization errors in noisy channels but will result in higher quantization errors than traditional vector quantization in noiseless channels.This paper presents a robust joint source and channel coding scheme that improves the robustness of the vector quantization in very noisy channels without sacrificing the quantization accuracy in noiseless channels.The scheme takes advantage of both COVQ and unequal error protection(UEP).This combination not only improves the COVQ performance in noiseless channels,but also enhances the anti-error capability in very noisy channels.Simulations in a speech coding system show that this scheme reduces the COVQ quantization error by about 50% in noiseless channels and improves the MOS performance by 0.58 over traditional source channel coding schemes in very noisy channels.更多还原