论文部分内容阅读
在背景噪声存在的情况下,电子耳蜗(cochlear implant,CI)的性能会受到严重影响,而传统谱减语音增强算法不能有效去除电子耳蜗中的各种非平稳噪声。为了进一步提高电子耳蜗使用者的听觉感知能力,本文提出了一种改进的谱减算法,其将带噪语音信号的功率谱进行Bark分带,在有声帧的每个子带内按照最小统计量控制递归平均方法进行噪声估计,并自适应地调节谱减参数。将该算法应用到电子耳蜗的前端预处理并仿真电子耳蜗合成声音。结果表明:相比传统的谱减法,该算法能够进一步提高合成语音的清晰度和可懂度。
In the presence of background noise, the performance of cochlear implant (CI) will be seriously affected, while the traditional spectral subtraction speech enhancement algorithm can not effectively remove various non-stationary noise in cochlear implants. In order to further improve the auditory perception ability of the cochlear user, an improved spectral subtraction algorithm is proposed in this paper. Bark banding is performed on the power spectrum of the noisy speech signal and is controlled according to the minimum statistics in each sub-band of the audio frame The recursive averaging method performs noise estimation and adaptively adjusts spectral subtraction parameters. The algorithm is applied to the front of the cochlear implant to pre-process and simulate the cochlear synthesis sound. The results show that this algorithm can further improve the clarity and intelligibility of synthesized speech compared with traditional spectral subtraction.