论文部分内容阅读
提出一种高质量的低速率语言编码算法ＭＷＩ，该算法对传统的波形插值算法ＷＩ进行了全面的改进。ＭＷＩ采用了一种更加合理的插值模型，包括统一的清音和浊音分析合成模型、新的典型波形的提取和表示方法，并且采用动态规划的算法增加基音周期估计的准确性。由于不依赖于准确的清浊音判别，ＭＷＩ具有更强的抗噪声能力。ＭＷＩ的参数设置也有利于更加有效的参数量化。随着编码速率的下降，ＭＷＩ的质量虽然呈下降的趋势，但在２ｋｂ／ｓ的速率下，它仍然能给出具有很高清晰度和自然度的合成语音。构造的完整的ＭＷＩ２ｋｂ／ｓ的声码器，具有很高的自然度和可懂度。
A high-quality low-rate speech coding algorithm MWI is proposed, which improves the traditional waveform interpolation algorithm WI. MWI employs a more reasonable interpolation model that includes a unified voice and voiced analysis synthesis model, a new representative waveform extraction and representation method, and a dynamic programming algorithm to increase the accuracy of the pitch period estimation. MWI is more robust against noise as it does not rely on accurate voiced / unvoiced discrimination. MWI parameter settings are also conducive to more effective parameter quantization. As the encoding rate drops, the quality of the MWI, though declining, still gives synthesized speech with high definition and naturality at a rate of 2kb / s. Construction of a complete MWI2kb / s vocoder, with a high degree of naturalness and intelligibility.