Yet Another Algorithm for Pitch Tracking

Abstract

In this paper, we present a pitch detection algorithm that is extremely robust for both high quality and telephone speech.The kernel method for this algorithm is the ''NCCF or Normalized Cross Correlation" reported by David Talkin [IJ.Major innovations include: processing of the original acoustic signal and a nonlinearly processed version of the signal to partially restore very weak FO components; intelIigent peak picking to select multiple FO candidates and assign merit factors; and, incotporation of highly robust pitch contours obtained from smoothed versions of low frequency portions of spectrograms. Dynamic programming is used to fmd the "best"pitch track among all the candidates, using both local and transition costs. We evaluated our algorithm using the Keele pitch extraction reference database as "ground truth" for both "high quality" and ' 'telephone'' speech. For both types of speech, the error rates obtained are lower than the lowest reported in the literature. low frequency spectrograms are shown in the middle panel (original signal) and bottom panel (absolute value signal). The two curves overlaid on each spectrogram are explained in detaillater. This figure illustrates that FO is much more prominent in the lower panel than the middle panel (also verified by comparison with TIMIT version of same sentence). Similar effects were noted for many other sample signals, some studio quality as well. The strategy adopted was to completely process 0-7803-7402-9/02/$17 .00 «:l2002 IEEE 1-361

Harmonic modification and data adaptive filtering based approach to robust pitch estimation

“…Different algorithms have been implemented in the time domain (Kasi and Zahorian 2002;Shimamura and Kobayashi 2001;Dogan and Mendel 1992; Abu-Shikhan and Deriche 1999) but none of them meets the desired performance of pitch estimation. The papers Kasi and Zahorian (2002), Shimamura and Kobayashi (2001) implement the autocorrelation function (ACF) based approaches which are popular for their simplicity and low computational complexity. The performance of ACF based algorithm is basically depended on the pitch peak in autocorrelation domain.…”

“…It is considered that there obviously remains one subband whose fundamental period (FP) is equal to the preliminary estimated pitch period. This reported pitch period is obtained with the help of WAC (Shimamura and Kobayashi 2001) or NACF (Kasi and Zahorian 2002) based method mentioned earlier. Then from the EMD domain, pitch periods of all subbands are estimated.…”

“…The actual pitch peak may be quite difficult to detect due to the presence of noise, quasi-periodic nature of the speech signals and the choice of analysis frames and widows (Hasan et al 2003(Hasan et al , 2005. A normalized autocorrelation function (NACF) based method has been introduced in Kasi and Zahorian (2002) which provides better pitch estimation accuracy than the ACF based method but it is not robust. Also the NACF based algorithm does not use efficient filtering techniques.…”

Empirical Mode Decomposition for Advanced Speech Signal Processing

“…There are many pitch estimation algorithms available now-a-days. Different algorithms have been implemented in the time domain [48,49] but none of them meets the desired performance of pitch estimation. The pitch estimation is also performed in the transformed domain.…”

FDMSM robust signal representation for speech mixtures and noise corrupted audio signals

“…In this respect the lack of an effective signal representation capable for analysis mixed signals is often introduced as a challenging problem in many applications. This difficulty is mainly due to that many pitch-dependent analysis methods including harmonic plus noise model (HNM) [1], are not capable of analyzing mixed signals since the performance of the state-of-the-art pitch estimation algorithms [2,3,4] may severely degrade by introducing large errors for mixtures [5]. The purpose of this study is to investigate the capability of the FDMSM signal representation for a wide range of audible signals including speech mixtures, songs, and noise corrupted audio signals.…”