The paper investigates the interdependence between the perceptual identification of the vocalic quality of six isolated Polish vowels traditionally defined by the spectral envelope and the fundamental frequency F0. The stimuli used in the listening experiments were natural female and male voices, which were modified by changing the F0 values in the ±1 octave range. The results were then compared with the outcome of the experiments on fully synthetic voices. Despite the differences in the generation of the investigated stimuli and their technical quality, consistent results were obtained. They confirmed the findings that in the perceptual identification of vowels of key importance is not only the position of the formants on the F1 × F2 plane but also their relationship to F0, the connection between the formants and the harmonics and other factors. The paper presents, in quantitative terms, all possible kinds of perceptual shifts of Polish vowels from one phonetic category to another in the function of voice pitch. An additional perceptual experiment was also conducted to check a broader range of F0 changes and their impact on the identification of vowels in CVC (consonant, vowel, consonant) structures. A mismatch between the formants and the glottal tone value can lead to a change in phonetic category.
Speech enhancement is fundamental for various real time speech applications and it is a challenging task in the case of a single channel because practically only one data channel is available. We have proposed a supervised single channel speech enhancement algorithm in this paper based on a deep neural network (DNN) and less aggressive Wiener filtering as additional DNN layer. During the training stage the network learns and predicts the magnitude spectrums of the clean and noise signals from input noisy speech acoustic features. Relative spectral transform-perceptual linear prediction (RASTA-PLP) is used in the proposed method to extract the acoustic features at the frame level. Autoregressive moving average (ARMA) filter is applied to smooth the temporal curves of extracted features. The trained network predicts the coefficients to construct a ratio mask based on mean square error (MSE) objective cost function. The less aggressive Wiener filter is placed as an additional layer on the top of a DNN to produce an enhanced magnitude spectrum. Finally, the noisy speech phase is used to reconstruct the enhanced speech. The experimental results demonstrate that the proposed DNN framework with less aggressive Wiener filtering outperforms the competing speech enhancement methods in terms of the speech quality and intelligibility.