摘要
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。
A new compression algorithm was developed to improve speech intelligibility in digital hearing aids. Morphology methods were used to localize the spectral cues, including formats and the spectral envelope. Gain values were interpolated to get the entire amplified spectrum. An inverse IFFT transformation was used to provide time-domain waveform output. Logical judgments and addition operations were used for spectral cues localization. As a result, the computations were dramatically reduced with simulation results indicating that the formats and spectral envelope were perfectly preserved. Primary speech intelligibility tests with severely hearing impaired people demonstrated that algorithm users achieve higher scores in consonant recognition tests and speech intelligibility tests in low noise circumstance than traditional LPC algorithm users.
出处
《清华大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2005年第12期1680-1683,共4页
Journal of Tsinghua University(Science and Technology)
关键词
数字助听器
形态学
压缩放大算法
谱特征
digital hearing aid
morphology
amplification and compression algorithm
spectral cues
