Magdalena Igras

Wavelet method for breath detection in audio signals

An algorithm for automatic detection of breath events in a speech signal is suggested in this paper. The issues of breath events occurrences in recordings are discussed as well as their statistical parameters. Also the role of breath pauses for signalizing punctuation and emotional or physical state of the speaker, in both spontaneous and read speech, is described. Wavelet parameters of energy in frequency subbands are obtained from discrete packet wavelet decomposition with mel-scale as patterns of breaths. In the beginning of the detection procedure, preliminary hypotheses of breath are indicated in the analyzed speech signal using temporal features. Then, discrete wavelet transform parameters are calculated. Dynamic time warping is applied to establish final recognition. The algorithm will be used for automatic speech recognition for measuring similarity between training and test features vectors.

Combination of Fourier and wavelet transformations for detection of speech emotions

The paper presents an approach to automatic recognition of emotions in speech signals. The applied method bases on the composition of two discrete frequency transformations. The wavelet transform was calculated first and next the Fourier transform was applied. The Fourier-wavelet transform representation is used to find the differences between emotions in speech signals. A set of approximately 30 seconds long speech signals was used to verify the efficiency of presented methods. It gives the possibility of analyzing the performance of speech emotion recognition in the Fourier-wavelet domain.

Is phoneme length and phoneme energy useful in automatic speaker recognition

The paper presents analysis of prosodic parameters of speech (energy, phoneme duration) as features characteristic for speaker. The most significant parameters of the features were investigated using CORPORA speech database and described statistically. We observed that phoneme duration depends on a speaker, as well as the preboundary lengthening of the phonemes in sentences. An average phoneme energy and an amount of energy per time are speaker-specific values also. These features can be used as complementary to a standard feature vectors of time-energy distributions for speaker recognition systems. The results of the investigation can be also applied to speech modeling for automatic speech recognition.

Looking for natural voice — The effectiveness of the program of Postgraduate Studies of Voice and Speech Training

The purpose of the study is to determine the acoustic aspects of the impact of a voice training intervention. The research group participated in a voice training program of Postgraduate Studies of Voice and Speech Training at the University of Social Sciences and Humanities in Warsaw. The subjects of the research are adults, aged 25-61, male and female. The initial recording session took place before the training cycle started. The recording of the same content was repeated to mark the stages of the training progress. The recordings were evaluated by an independent expert (perceptual evaluation, using e.g. the GRBAS scale) and by the participants themselves (subjective self-evaluation, Voice Handicap Index questionnaire). Furthermore, a case study of each course participant was conducted. As an objective assessment technique, acoustic analysis of the collected speech signals was performed. A statistical analysis shows which parameters correspond to the perceptual categories of vocal performance evaluation. The most significant parameters are chosen to develop algorithms in order to track progress within training course. For the analysis, Praat and Matlab software was used. The study objectively confirms the changes in voice quality of the participants during the study period.

Caller identification by voice

The aim of our work is to develop the software for caller identification or to create his characteristic by analysis of his voice. Based on collected speech samples, our system aims to identify emergency callers both on-line and off-line. This homeland security project covers speaker recognition (when speakers speech sample is known), speakers gender, age detection and recognition of emotions. Proposed system is not limited to bio-metrics. The goal of this application is to provide an innovative, supporting tool for rapid and accurate threat detection and threat neutralization. This complex system will include: a speech signal analysis, an automatic development of speech patterns database and appropriate classification methods.

VOCAL TRACT NORMALISATION IN COMPUTER GAMES

A new approach to speech normalisation is presented. A method that finds the optimal coefficient for linear slope of the warping function is described. The affine normalisation functions are suggested. Their coefficients depend on expected values of frequency when speech spectra are used as a density of probabilities. The method was developed for computer games to lower costs of recording dialogues and to make them more attractive for players.