基于统计特征和显著度的光谱信号提取算法

各类光谱信号都会受到噪声和基线畸变的影响,在提取光谱信号过程中若不考虑基线畸变和噪声的影响,将会严重影响信号提取的精度和准确性,所以需要在信号提取前消除噪声和基线畸变的影响.大多数信号提取算法的步骤是先提取整体基线,再提取信号,这样难以保证基线的提取精度.为了降低信号提取过程中背景噪声、基线畸变等不利因素的影响,根据信...     

兴奋模式下谐波复合音音高感知机制的研究

通过测量谐波复合音的基频辨别阈,探讨中等\     

Robust and accurate fundamental frequency estimation based on dominant harmonic components

This paper presents a new method for robust and accurate fundamental frequency (F0) estimation in the presence of background noise and spectral distortion. Degree of dominance and dominance spectrum are defined based on instantaneous frequencies. The degree of dominance allows one to evaluate the magnitude of individual harmonic components of the speech signals relative to background noise while reducing the influence of spectral distortion. The fundamental frequency is more accurately estimated from reliable harmonic components which are easy to select given the dominance spectra. Experiments are performed using white and babble background noise with and without spectral distortion as produced by a SRAEN filter. The results show that the present method is better than previously reported methods in terms of both gross and fine F0 errors.     

Pitch circularity from tones comprising full harmonic series

This paper describes an algorithm for producing pitch circularity using tones that each comprise a full harmonic series, and reports an experiment that demonstrates such circularity. Banks of 12 tones (i.e., scales) were created, with F0 varying in semitone steps. For each scale, as F0 descended, the amplitudes of the odd-numbered harmonics were reduced relative to the even-numbered ones by 3.5 dB for each semitone step. In consequence, the tone with the lowest F0 was heard as though displaced up an octave. In an experiment employing two such scales, all possible ordered tone pairs from each scale were presented, making 132 ordered tone pairs for each scale. Sixteen subjects judged for each tone pair whether the second tone was higher or lower than the first. The data derived from these pairwise comparisons were subjected to Kruskal's nonmetric multidimensional scaling, and excellent circularities were obtained. Individual differences in the subjects' judgments were also explored. The findings support the argument that musical pitch should be characterized as varying along two dimensions: the monotonic dimension of pitch height and the circular dimension of pitch class.     

Pitch perception of concurrent harmonic tones with overlapping spectra

Fundamental frequency difference limens (F0DLs) were measured for a target harmonic complex tone with nominal fundamental frequency (F0) of 200 Hz, in the presence and absence of a harmonic masker with overlapping spectrum. The F0 of the masker was 0, ± 3, or ± 6 semitones relative to 200 Hz. The stimuli were bandpass filtered into three regions: 0-1000 Hz (low, L), 1600-2400 Hz (medium, M), and 2800-3600 Hz (high, H), and a background noise was used to mask combination tones and to limit the audibility of components falling on the filter skirts. The components of the target or masker started either in cosine or random phase. Generally, the effect of F0 difference between target and masker was small. For the target alone, F0DLs were larger for random than cosine phase for region H. For the target plus masker, F0DLs were larger when the target had random phase than cosine phase for regions M and H. F0DLs increased with increasing center frequency of the bandpass filter. Modeling using excitation patterns and "summary autocorrelation" and "stabilized auditory image" models suggested that use of temporal fine structure information can account for the small F0DLs obtained when harmonics are barely, if at all, resolved.     

基于谐波和噪声能量改进的语音截止频率轮廓估计

为了对语音谐波/噪声模型中的语音截止频率轮廓进行更好的描述,本文提出了一种基于谐波和噪声能量改进的语音截止频率轮廓估计算法。改进算法对累积谐波和噪声能量函数进行对应谐波处的功率谱加权,并且在语音截止频率轨迹的平滑部分采用形态滤波的手段。实验表明,与原算法相比,通过改进算法得到的截止频率轮廓在语谱图上标注得更为准确,在主观评分测试中改进算法也获得了优于原算法的测试评分。改进算法比原算法能够更加准确的对语音截止频率轮廓进行估计,从而使得语音谐波/噪声模型在语音编码、语音合成及识别方面具有更为有效的应用。     

Pitch estimation by early-deafened subjects using a multiple-electrode cochlear implant

Numerical estimates of pitch for stimulation of electrodes along the 22-electrode array of the Cochlear Limited cochlear implant were obtained from 18 subjects who became deaf very early in life. Examined were the relationships between subject differences in pitch estimation, subject variables related to auditory deprivation and experience, and speech-perception scores for closed-set monosyllabic words and open-set Bamford-Kowal-Bench (BKB) sentences. Reliability in the estimation procedure was examined by comparing subject performance in pitch estimation with that for loudness estimation for current levels between hearing threshold and comfortable listening level. For 56% of subjects, a tonotopic order of pitch percepts for electrodes on the array was found. A deviant but reliable order of pitch percepts was found for 22% of subjects, and essentially no pitch order was found for the remaining 22% of subjects. Subject differences in pitch estimation were significantly related to the duration of auditory deprivation prior to implantation, with the poorest performance for subjects who had a longer duration of deafness and a later age at implantation. Subjects with no tonotopic order of pitch percepts had the lowest scores for the BKB sentence test, but there were no differences across subjects for monosyllabic words. Performance in pitch estimation for electrodes did not appear to be related to performance in the estimation procedure, as all subjects were successful in loudness estimation for current level.     

多谐波脉冲星信号时延估计方法

针对脉冲星导航技术中延时估计这一关键问题, 提出了频域上直接使用脉冲星信号测量到达时间集合进行时延估计的方法——多谐波脉冲星信号时延估计(MHSPE)方法. 该方法建立在频域上相位时延的极大似然估计的基础上, 通过高次谐波对脉冲星观测信号提取出各谐波相位的极大似然估计, 然后取频谱上各谐波的幅值进行归一化作为各谐波相位的权值, 最后取各谐波相位的加权平均作为该时刻的相位估计. 理论上证得MHSPE算法对相位的估计是无偏、一致的, 相比于频域上一次谐波的极大似然估计, MHSPE方法的信噪比随谐波数m的增加而增加, 当各谐波幅值相同时, 信噪比可提高m_1/2倍; 与脉冲星信号时延的克拉美罗界比较, 脉冲星信号时域的导数在频域上的反映就是各谐波分量的数量, 因此随着谐波次数的增加脉冲星信号时延估计可极大趋近克拉美罗界. 采用RXTE航天器对Crab脉冲星的实测数据检验MHSPE方法的性能, 实验结果表明, 针对低信噪比的脉冲星信号, MHSPE可获得高精度的相位估计, 随观测时间增加, 估计精度快速收敛于克拉美罗界.     

Pitch discrimination of harmonic complex signals: residue pitch or multiple component discriminations?

Two models for pitch discrimination of harmonic complex sounds are discussed, a multiple-band probability summation model using comparisons among component frequencies, and a model in which residue pitches are compared. The second model is based on Goldstein's optimum-processor pitch theory [J. Acoust. Soc. Am. 54, 1496-1516 (1973)], and is distinguished from the multiple-band model by an internal noise process. Pitch difference limens from 2I2AFC tasks show that when the test signals comprise corresponding harmonics, relative pitch difference limens are less than the smaller relative difference limens for the component frequencies, which is consistent with the multiple-band model. The absence of corresponding harmonics significantly reduces relative pitch discriminability; this effect supports the model on Goldstein's theory. It appears that residue pitch comparisons are not used for pitch discrimination between sounds with corresponding components; rather, comparisons based on residue pitch are only employed where there are no common resolved components in the signals to be discriminated.     

基于数学形态滤波的语音信号基音特征提取

数学形态滤波是一种关于信号形状处理的非线性变换,它能简化信号、消除较小分量而保留信号的基本形状特征.本文基于数学形态滤波方法提出了两个分别在时域和频域提取语音信号基音周期的方案,在频域提取基音周期的同时还能提取出语音信号的谱包络。它们具有简单、直观和计算效率高等特点。由于数学形态滤波运算是并行的、局部的,新方案适于并行化处理和易于硬件化实现。实验结果表明,选择合理的数学形态滤波参数以及线性预测编码参数,能获得准确的语音信号基音特征。     

基于Hilbert-Huang变换的基音周期提取方法

提出了一种基于Hilbert-Huang变换的语音信号基音周期提取方法。该方法无须对语音信号进行分帧截断。语音信号直接进行Hilbert-Huang变换后,通过基音频率搜索处理得到基音频率及其随时间的变化。实验结果表明,与传统的基音周期提取方法相比,该方法既能真实描述语音信号的非平稳非线性特性,又能提高基音周期提取的准确性和分辨率。