Fundamental frequency estimation using signal embedding in state space

Abstract

A new robust nonlinear method for determination of fundamental frequency (F0) was recently proposed with application to speech pitch detection [Terez, Proc. ICASSP 1, 345–348 (2002)]. The method uses state-space embedding technique originally introduced for analyzing chaotic signals. The new method has been generalized and tested on different types of speech signals, as well as on a variety of other acoustic signals. In addition, some artificially generated nonstationary and complex wave forms have been used to test the limits of the method in comparison with other known (short-term) F0-estimation techniques (e.g., correlation, spectrum or cepstrum-based methods). Evaluation results demonstrate a unique combination of properties distinguishing the new method from conventional techniques. In particular, reliable and accurate F0 estimates can be obtained for clean periodic signals using signal segments slightly longer than one complete fundamental period. Other properties include immunity to speech formants and robust performance on noisy and band-limited speech signals. Some improvements are introduced to reduce the number of required computations and to achieve higher (subsample) accuracy. The method waused to implement a robust pitch-tracking algorithm for speech processing applications. Further information and demo software can be found at http://www.soundmathtech.com/pitch.

Text-Independent Speaker Identification System Using Average Pitch and Formant Analysis

“…Pitch represents the perceived fundamental frequency (F0) of a sound and is one of the major auditory attributes of sounds along with loudness and quality [13][14]. Here we are interested to find out the average pitch of a speech signal.…”

Relative-pitch tracking of multiple arbitrary sounds

“…This is a problem which has been tackled using a rich variety of approaches and continues to inspire a considerable amount of research. Approaches to pitch track extraction have ranged from straightforward period estimation, to sophisticated statistical methods, some employing time domain techniques and others sophisticated front-ends that reveal more of the pitch structure [1][2][3][4][5][6][7][8]. The applications of pitch tracking cover a wide range of applications ranging from musical transcription, to emotion recognition in speech, to animal acoustics.…”