CAJ | 학술논문

为了降低复杂环境中存在的各类噪声,消除噪声给语音识别及说话人识别等系统在军事自动指挥及控制领域等的应用中带来的影响,设计了一种新的谱减法。针对传统谱减法在语音增强过程中的原始语音和噪声信号相差为0的错误假设,提出一种基于复平面的、考虑纯净语音与噪声及带噪语音之间相位差信息的改进谱减法,对增益函数进行修正。经过复杂的战场仿真环境测试表明,该方法性能较好,在低信噪比的环境下性能更优。对于输入信噪比为5 dB的带噪语音,经复平面谱减法去噪后,输出信噪比为12.21 dB,比传统谱减法提高了2.1 dB。
위료강저복잡배경중존재적각류조성,소제조성급어음식별급설화인식별등계통재군사자동지휘급공제영역등적응용중대래적영향,설계료일충신적보감법。침대전통보감법재어음증강과정중적원시어음화조성신호상차위0적착오가설,제출일충기우복평면적、고필순정어음여조성급대조어음지간상위차신식적개진보감법,대증익함수진행수정。경과복잡적전장방진배경측시표명,해방법성능교호,재저신조비적배경하성능경우。대우수입신조비위5 dB적대조어음,경복평면보감법거조후,수출신조비위12.21 dB,비전통보감법제고료2.1 dB。
To reduce noise in complicated environments and eliminate the inferences of noise on the applications of speech recognition and speaker recognition systems in automatic military command and control fields,a new spectral subtraction is designed. Aiming at the incorrect assumption of the traditional spectral subtraction that the phase difference between the clean speech and noise signals is zero in the speech enhancement process, an improved spectral subtraction is proposed based on complex plane and considering the phase difference among the clean speech,the noise and the noisy speech signals. The gain function is corrected. The results of complicated battlefield simulation experiments show that the method proposed here performs better than the traditional spectral subtraction,especially in low signal noise ratio ( SNR) conditions. For the noisy speech with input SNR of 5 dB,the output SNR is more than 12. 21 dB after denoised by using the complex plane spectral subtraction,which increases 2. 1 dB than the result of the traditional spectral subtraction.