比索
语音增强
计算机科学
语音识别
光谱图
可理解性(哲学)
光谱密度
噪音(视频)
噪声功率
失真(音乐)
噪声测量
振幅
算法
背景噪声
人工智能
降噪
功率(物理)
带宽(计算)
电信
物理
哲学
图像(数学)
放大器
认识论
量子力学
作者
Qiujuan Tong,Huan Han,Lu Huang,Jun Wang
标识
DOI:10.1145/3488933.3488984
摘要
In traditional speech enhancement algorithms, the amplitude of noise spectrum is only modified, while the phase and power spectral density estimation of noise power spectrum is inaccurate, resulting in serious distortion of synthesized speech signal. Aiming at addressing this problem, a speech enhancement algorithm combining amplitude estimation and phase reconstruction is proposed in this paper. First, we use the minimum controlled recursive averaging (MCRA) function to estimate the power spectrum of the noise, instead of relying only on the initial silent frame to achieve the amplitude estimation of the enhanced signal. Secondly, the phase spectrum of noisy speech signal is reconstructed. Finally, the reconstructed phase spectrum is combined with the estimated speech amplitude spectrum to generate clean speech signal. In order to evaluate the performance of the proposed algorithm, the NOIZEUS data set is used for simulation. The objective experiments of perceptual estimation of speech quality (PESQ), normalized covariance (NCM), and spectrogram analysis show that the proposed method can suppress noise signals greatly, reduce signal distortion at low signal-noise ratio (SNR), and improve speech quality and intelligibility.
科研通智能强力驱动
Strongly Powered by AbleSci AI