Unified Coding Method Of Speech And Audio Signals

With the rapid development of network communication, mobile communicationand multimedia technology, the mutual fusion between the different networks,systems and service platforms has become inevitable. Under this trend, the gapbetween the communication and entertainment has been fading away. People cannotbe satisfied by the monotonous voice communication, the enjoyment of both speechand audio communication is highly desired. But the traditional speech and audiocodec cannot deal equally well with all types of audio content including both speechand music. This has hindered the further development of multimedia technology.Under this background, the Moving Picture Expert Group (MPEG) issued a Callfor Proposal of Unified Speech and Audio Codec (USAC). The desired codec canperform equally well on all types of audio content by using one unified codingalgorithm, which cannot be achieved by any traditional speech or audio codecindependently. Because of great potential of USAC, a lot of institutions are engagedinto the research of unified coding method at present.In order to solve this problem, the work of this dissertation is just focus onconstructing a unified speech and audio coding method. After researching the keytechnologies of the existing speech and audio codecs, several unified codingalgorithms that based on the common harmonic characters of speech and audio signalare proposed in this dissertation. Finally, a wide-band unified speech and audio codecis proposed at the bitrate of24kbps and32kbps.The main research contributions are as follows:1. Based on the thought of characteristic components separation, a unified codingalgorithm by preserving the possibility distribution of original signal is proposed. Theproposed codec treats the input speech and audio as a general signal. So the distortionthat introduced by the misjudgment of input signal type and the mode switch can beeliminate effectively.2. An adaptive harmonic separation algorithm based on the Empirical ModeDecomposition (EMD) is proposed in this paper. Under the constraint of perceptualsensitivity and periodicity of the Intrinsic Mode Functions (IMF), the harmonics oforiginal signal is extracted by using the adaptive filter characteristic of EMD. Afterthis, the accuracy of sinusoidal model parameters estimation is improved. 3. A sinusoidal parametric unified codec based on harmonic separation isproposed in this paper. In order to exploit the advantages of both parametric codingand transform coding, a hybrid coding architecture is applied to different componentsof original signal. For harmonic components, a gradient perceptual weighted matchingpursuit is used for sinusoidal modeling and multiresolution quantization; fornoise-like components, a dithered RE8lattice quantization is applied to modulate thequantization noise into a white noise.4. For the further improvement of the proposed sinusoidal codec, the pitchsynchrony analysis combined with power spectral preserving quantization is adoptedto improve the quality of speech signal coding. The proposed pitch synchrony speechquantizer warp the input signal into a constant pitch period signal firstly, and then theenergy of modulation transform coefficients of voiced speech is concentrated intolowered modulation bands by using a sparse transformation.5. Based on the pitch synchrony analysis method, an adaptive analysis windowlength judgment algorithm by using the normalized energy weighted correlationcoefficients is adopted for the analysis of both speech and audio signal. By combiningwith the Distribution Preserving Quantization (DPQ), a unified speech and audiocodec based on DPQ is proposed. Both the objective PESQ/PEAQ results and thesubjective A/B listening tests show that the proposed coder outperforms the ITU-TG.722.1and AMR-WB codec.