Design Of Speech Synthesis Based On AMR-WB

With rapid development of the computer and information technology, the speech synthesis technology has become the hot topic which the scientific researchers pay attention to. Linear prediction coding (LPC) synthesis has been gained special attention in speech synthesis for its lower rate, lower complexity, lower cost, simple and intuitive characteristics. However, Because of the parameter extraction of original model is too simple,it is not easy to obtain high-quality synthesis speech in the synthesis stage. Adaptive Multi-Rate Wideband (AMR-WB) speech coding technology is a high-quality digital wideband speech coding system. In the parameter extraction stage, the Algebraic Code Excited Linear Prediction Coder (ACELP) is adopted, the bandwidth is broadened from traditional 200-3400Hz to 50-7000Hz, the combination of closed-loop search and open-loop search method is used to calculate the pitch delay and the combination of adapted codebook and fixed codebook is used to retrieve excitation signal. Those improve the synthesis effect and provide a very good thought for improving the effect of LPC parameter speech synthesis.In this paper, after researching on the AMR-WB codec algorithm in detail, Chinese TTS system based on AMR-WB has been designed. In the parameter extraction stage, input sampling signal with 7k bandwidth, 16 kHz frequency, and 16 bits precision, Compared with the input signal of standard LPC algorithm, has a higher voice quality. The combining method of closed-loop search and open-loop search is used as algorithm in the calculation of Pitch, and the combination of adaptive codebook and fixed codebook is adopted to retrieve excitation signal. These all have made up the standard LPC algorithm deficiency, and greatly improve the synthetic voice effect. To improve operating efficiency of parameter extraction, a procedure of conversing PCM is designed. In synthesis stage, according to the needs of TTS system, and considering input forms of Chinese character and punctuation, the synthesis programme is designed, which could obtain output of female voice or male voice according to the control command and manifest the punctuation information with a section of blank pronunciation. In verification stage, the synthesis results are simulated by using MATLAB software. The simulation results show that synthetic voice greatly represent the amplitude-frequency characteristics of voice samples, and obtained high quality. And the algorithm complexity analysis also proves that the improved algorithm meets requirements. So, it is feasible to use AMR-WB as an improved algorithm to improve the effect of LPC parameter speech synthesis.