论文部分内容阅读
为改善数字化助听器产品的语言识别,提出了一种单频段压缩放大方法。该方法利用数学形态学方法定位共振峰和频谱包络特征,通过对特征点处增益值进行插值的方法得到放大后的频谱,再利用Fourier变换实现单频段压缩放大。该算法中谱特征定位只需逻辑判断和加法即可实现,其计算量低于同类线性预测编码(LPC)算法。仿真结果表明,该算法对声谱特征点定位完整、准确,放大后谱特征保存完好。言语测试结果表明,该方法相对于传统的多频段压缩放大方法,可提高中重度聋患者的辅音识别效果和噪声背景下的言语识别率,且减小计算量,有实用价值。
To improve the speech recognition of digital hearing aids, a single-band compression amplification method is proposed. This method uses the method of mathematical morphology to locate the resonance peak and the spectrum envelope feature, obtains the amplified spectrum by interpolation of the gain value at the feature point, and then uses the Fourier transform to realize the single-band compression and amplification. The spectral feature location in this algorithm can be realized only by logical judgment and addition, which is lower than the linear prediction coding (LPC) algorithm. The simulation results show that this algorithm can locate the feature point of the sound spectrum completely and accurately, and preserve the spectral features well after amplification. Speech test results show that compared with the traditional multi-band compression amplification method, this method can improve the consonant recognition effect and the speech recognition rate under the noisy background in moderate to severe deaf patients, and reduce the computational burden.