A computationally efficient model for multipitch and periodicity analysis of complex audio signals is presented. The model essentially divides the signal into two channels, below and above 1000 Hz, computes a "generalized" autocorrelation of the low-channel signal and of the envelope of the high-channel signal, and sums the autocorrelation functions. The summary autocorrelation function (SACF) is further processed to obtain an enhanced SACF (ESACF). The SACF and ESACP representations are used in observing the periodicities of the signal. The model performance is demonstrated to be comparable to those of recent time-domain models that apply a multichannel analysis. In contrast to the multichannel models, the proposed pitch analysis model can be run in real time using typical personal computers. The parameters of the model are experimentally tuned for best multipitch discrimination with typical mixtures of complex tones. The proposed pitch analysis model may be used in complex audio signal processing applications, such as sound source separation, computational auditory scene analysis, and structural representation of audio signals. The performance of the model is demonstrated by pitch analysis examples using sound mixtures which are available for download at http://www.acoustics.hut.fi/-ttolonen/pitchAnalysis/.

/pdf/a-computationally-efficient-multipitch-analysis-model-4lkkljkndh.pdf

A computationally efficient multipitch analysis model

Based on a new similarity model for the voice excitation process, a novel pitch determination procedure is derived. The unique features of the proposed algorithm are infinite (super) resolution, better accuracy than the difference limen for F/sub 0/, robustness to noise, reliability, and modest computational complexity. The algorithm is instrumental to speech processing applications which require pitch synchronous spectral analysis. The computational complexity of the proposed algorithm is well within the capacity of modern digital signal processing (DSP) technology and therefore can be implemented in real time. >

Super resolution pitch determination of speech signals

We present a two-channel technique for improving speech analysis in certain applications. One channel is the signal from an electroglottograph (EGG), which monitors the vibratory motion of the vocal folds. The other channel is the speech signal obtained from a conventional microphone. We show how the EGG can be used as a tool for validating speech processing algorithms and estimating possible lower bounds for both computation and performance of these algorithms, particularly closed-phase speech analysis. Our system is used to classify speech segments as voiced, unvoiced, mixed voiced, and silent and to estimate the fundamental frequency of voicing. This four-way classification is not implemented as a complete algorithm; it still requires some user judgments and decisions. The technical results, however, illustrate an EGG-based algorithm for voiced/unvoiced-silent classification. In addition, we illustrate how automatic on-line inverse filtering can be achieved. The results demonstrate the superiority of the closed-phase covariance analysis method over several other commonly used methods. Source-tract coupling is shown to be a significant factor in linear prediction analysis, a factor commonly ignored to date. Various applications of our two-channel approach are described along with the major disadvantage, namely, that in some situations the EGG channel cannot be acquired.

Two-channel speech analysis

A method is described for estimating the fundamental frequencies of several concurrent sounds in polyphonic music and multiple-speaker speech signals. The method consists of a computational model of the human auditory periphery, followed by a periodicity analysis mechanism where fundamental frequencies are iteratively detected and canceled from the mixture signal. The auditory model needs to be computed only once, and a computationally efficient strategy is proposed for implementing it. Simulation experiments were made using mixtures of musical sounds and mixed speech utterances. The proposed method outperformed two reference methods in the evaluations and showed a high level of robustness in processing signals where important parts of the audible spectrum were deleted to simulate bandlimited interference. Different system configurations were studied to identify the conditions where pitch analysis using an auditory model is advantageous over conventional time or frequency domain approaches.

Multipitch Analysis of Polyphonic Music and Speech Signals Using an Auditory Model

Variation of electrolaryngographically derived closed quotient for trained and untrained adult female singers

An algorithm is presented which is designed to yield reference pitch contours of speech signals for a comparative evaluation of pitch determination algorithms. For the task of pitch determination the question of error analysis in general and measurement accuracy in particular is discussed, It is shown that the accuracy of the algorithm must be better than 0.5% in order to render fine measurement errors inaudible. The algorithm uses the output signal of a laryngograph which measures the laryngeal vibrations via the changes of the electric impedance of the larynx. The point of inflection during the rapid rise of the laryngogram is taken as the estimate for the instant of glottal closure. Using a local interpolation filter the quantization error of the measurement is kept below 0.5%.

Accurate pitch determination of speech signals by means of a laryngograph

The performance of double-transform pitch determination algorithms (PDAs) with frequency-domain nonlinear distortion is investigated. This PDA principle comprises well-known PDAs, such as the autocorrelation PDA and the cepstrum PDA. Besides these two, the amplitude spectrum and the fourth-root spectrum PDAs were implemented and tested under various environmental conditions (clean signal, telephone-quality signal, and signal degraded by additive Gaussian noise). First results suggest that the fourth-root spectrum and the amplitude spectrum PDAs are less sensitive to noise than the cepstrum PDA and less sensitive to strong first formants than the autocorrelation PDA.

H. Indefrey

Papers

Accurate pitch determination of speech signals by means of a laryngograph

Design and evaluation of double-transform pitch determination algorithms with nonlinear distortion in the frequency domain-preliminary results