Instantaneous RAPT algorithm

Abstract

This is the homepage of instantaneous RAPT (IRAPT) algorithm. It is a pitch estimation technique based on the robust algorithm for pitch tracking (RAPT) framework. IRAPT provides estimation of instantaneous pitch values and is not sensitive to rapid frequency modulations. The algorithm utilizes a different period candidate generating function based on instantaneous harmonic parameters. The function is similar to normalized cross-correlation function, however it represents momentary periodicity with high frequency resolution. The second major revision of RAPT is an additional post-processing procedure that makes estimation values more accurate. A short algorithm description is given in
Azarov, E., Vashkevich, M. and Petrovsky, A., "Instantaneous pitch estimation based on RAPT framework" Proceedings of the 20th European Signal Processing Conference (EUSIPCO'2012), pp. 2787-2791, Bucharest, Romania, August 27-31, 2012.

Matlab implementation of IRAPT algorithm can be found here .

You are welcome to send any questions, suggestions or error reports to the authors Elias Azarov azarov@bsuir.by and Maxim Vashkevich vashkevich.m@gmail.com.

Introduction

A pitch estimation technique based on the robust algorithm for pitch tracking (RAPT) framework [1]. The main features of the proposed algorithm are:

The main ideas behind the algorithm are:

Pitch estimation scheme

The pitch estimation scheme provides two pitch estimations: initial pitch estimation (denoted as "IRAPT 1") and recalculated pitch in time warped domain (denoted as "IRAPT 2").

Estimation of instantaneous harmonic parameters

The signal is decomposed into overlapping bandlimited analytical signals :

where - bandwidth and - center frequency of the i-th band and - an even window function. Then instantaneous parameters are evaluated as

where and are real and imaginary parts of respectively.

Period candidate generating function

RAPT
Normalized cross-correlation function (NCCF)

where and n - window size.
IRAPT
NCCF is calculated using instantaneous harmonic parameters

where P is the number of bandlimited analytical signals.

Time warping and pitch recalculation

Time warping eliminates pitch modulations. The signal is sampled with variable frequency aligned with estimated initial pitch contour. Resampling is carried out using sinc interpolation.

Pitch recalculation: new pitch values (denoted as "IRAPT 2" on the scheme) are calculated as:

Experimental results

The proposed technique is compared with other pitch estimation algorithms in terms of gross pitch error (GPE, %) and mean fine pitch error (MFPE, %).

Artificial signals

Natural speech

 

Male

Female

GPE

MFPE

GPE

MFPE

RAPT [1]

3.69

1.74

6.07

1.18

YIN [2]

3.18

1.39

3.96

0.84

SWIPE'[3]

0.78

1.51

4.27

0.80

IRAPT 1

1.63

1.61

3.78

0.98

IRAPT 2

1.57

1.57

3.78

1.05

Conclusions

An instantaneous pitch estimation algorithm has been proposed. The algorithm has low sensitivity to rapid pitch modulations and provides accurate values with high time/frequency resolution. Due to its properties the algorithm can be useful in various applications such as speech analysis/synthesis and automatic rotational frequency tracking for mechanical devices.

Reference

[1] D. Talkin, "A Robust Algorithm for Pitch Tracking (RAPT)" in "Speech Coding & Synthesis", W B Kleijn, K K Paliwal eds, Elsevier ISBN 0444821694, 1995.

[2] A. Cheveigne and H. Kawahara "YIN, a fundamental frequency estimator for speech and music", Journal Acoust. Soc. Am., vol. 111, no. 4, pp 1917-1930, Apr. 2002.

[3] A. Camacho and J. G. Harris, "A sawtooth waveform inspired pitch estimator for speech and music", Journal Acoust. Soc. Am., vol. 123, no. 4, pp 1638-1652, Sep. 2008.

Copyrights to these papers may be held by the publishers. The download files are preprints. All persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.