A comparative performance study of several pitch detection algorithms

Abstract

Abstract-A comparative performance study of seven pitch detection algorithms was conducted. A speech data base, consisting of eight utterances spoken by three males, three females, and one child was constructed. Telephone, close talking microphone, and wideband recordings were made of each of the utterances. For each of the utterances in the data base; a "standard" pitch contour was semiautomatically measured using a highly sophisticated interactive pitch detection program. The "standard" pitch contour was then compared with the pitch contour that was obtained from each of the seven programmed pitch detectors. The algorithms used in this study were 1) a center clipping, infinite-peak clipping, modified autocorrelation method (AUTOC), 2) the cepstral method (CEP), 3) the simplified inverse filtering technique (SIFT) method, 4) the parallel processing time-domain method (PPROC), 5) the data reduction method (DARD), 6) a spectral flattening linear predictive coding (LPC) method, and 7) the average magnitude difference function (AMDF) method. A set of measurements was made on the pitch contours to quantify the various types of errors which occur in each of the above methods. Included among the error measurements were the average and standard deviation of the error in pitch period during voiced regions, the number of gross errors in the pitch period, and the average number of voiced-unvoiced classification errors. For each of the error measurements, the individual pitch detectors could be rank ordered as a measure of their relative performance as a function of recording condition, and pitch range of the various speakers. Performance scores are presented for each of the seven pitch detectors based on each of the categories of error.

SpeakerSense: Energy Efficient Unobtrusive Speaker Identification on Mobile Phones

“…The speaker identification stage is based on a Gaussian Mixture Model (GMM) classifier [13] with some modifications to improve robustness and computation efficiency for use on the mobile phone. We use as our feature vector pitch [11] and the Mel-frequency cepstral coefficients (MFCCs) [20] computed for each admitted frame. SpeakerSense computes 20-dimensional MFCCs, and then ignores the first coefficient, which represents the energy of the frame, and instead focuses on the spectral shape, represented by the 2 nd through 20 th coefficients.…”

“…Similar to both of these systems, our goal is not to design new speaker identification algorithms. Instead we leverage well-established techniques such as the MFCCs feature set [20], pitch tracking [11], and GMM classifiers [e.g., 13,14], which have been proven effective for speaker identification. Our focus is on adapting these techniques to a mobile platform and addressing challenges that arise when using speaker identification on energy constrained mobile phones.…”

Extended Fundamental Frequency Extraction Using Exponentiated Amplitude Spectrum with Band-Limitation

“…All The fundamental frequency extraction is an important task in many speech processing systems [1] [2]. Therefore, extracting the fundamental frequency of a speech signal is essential for research in speech processing and many methods have been proposed [1]- [6].…”

“…Therefore, extracting the fundamental frequency of a speech signal is essential for research in speech processing and many methods have been proposed [1]- [6]. However, a performance improvement in noisy environments is still desired.…”

“…The autocorrelation function (ACF) [4] method is classified into this category, and behaves robustly in noisy environments [1]. However, the ACF has the property of being sensitive to the formant characteristic of vocal tract.…”