Speaker gender identification based on combining linear and nonlinear features

Abstract

Automatic speaker gender identification based on the speech feature has important application in the audio processing and analysis field. In order to overcome the conventional linear parameters in the speaker feature lack of gender characteristics, in this paper, nonlinear parameters such as the fractal dimension and fractal complexity as feature space effective compensations are presented. Firstly, use lifting scheme to extract pitch; Then extract the speech fractal dimension; Finally, according Takens theorem, time delay method is used to reconstruct phase space of fractal dimension sequence, fractal dimension complexity is obtained by calculating Approximate Entropy. Three dimension feature vectors constructed by the pitch, the fractal dimension and the fractal dimension complexity are applied to speaker gender identification. Results show as the identification system based on the new method introduces the non-linear parameters, its accuracy and stability are effectively improved compared with the traditional linear method identification systems. The new nonlinear method provides new ideas for speaker gender identification of a new line of thought.

Gender-Driven Emotion Recognition Through Speech Signals For Ambient Intelligence Applications

“…Other employed features are formant frequencies and bandwidths, open quotient and source spectral tilt correlates [12], energy between adjacent formants [15], fractal dimension and fractal dimension VOLUME 1, NO. 2, DECEMBER 2013 complexity [13], jitter and shimmer (pitch and amplitude micro-variations, respectively), harmonics-to-noiseratio, distance between signal spectrum and formants [16].…”

“…• Mean, variance, minimum, maximum and range of the first 12 Mel-Frequency Cepstrum Coefficients [13], [14];…”

“…As reported in papers such as [10], [11], [13], [32], [33] audio-based Gender Recognition (GR) has many applications. For example: gender-dependent model selection for the improvement of automatic speech recognition and speaker identification, content-based multimedia indexing systems, interactive voice response systems, voice synthesis and smart human-computer interaction.…”

“…Different kinds of classifiers are used to identify the speaker gender starting from features: e.g., Continuous Density Hidden Markov Models [13], [16], Gaussian Mixture Model (GMM) [14], [16], Neural Networks [10], [32], Support Vector Machines [12], [33]. The percentages of correct recognition of the speaker gender are reported in Table 1 for most classifiers referenced above.…”