Abstract—The speech fundamental frequency (F0) is an
important speech parameter which has significant effect in
domains such as voiceprint analysis, speaker recognition, voice
conversion and etc. The existing extraction methods of speech
fundamental frequency are not satisfactory in terms of accuracy
due to the limitations of the methods themselves and noise so
that correction is essential for high-quality analysis, recognition,
conversion. The effect of existing correction methods rely
heavily on the accuracy of first extraction, and are sensitive to
noise. This article proposes a new correction method which
makes use of cross-correlation of original signal and estimated
signal to convert the difference between their frequencies to
phase difference, thereby gains the accurate fundamental
frequency by measuring this phase difference. Experiments
indicate that the proposed method is effective when the
estimated frequency ranges from 70% to 190% of the actual
frequency, and keeps the average error within -50dB (0.35%)
when signal to noise ratio (SNR) reduces to 3dB.
Index Terms—Speech fundamental frequency, precision
correction, large scale, anti-noise, phase difference.
The authors are with the School of Electronic Science and Engineering,
Nanjing University, Nanjing, China (e-mail: zhaomangzheng@ gmail.com,
shining@nju.edu.cn, geyun@nju.edu.cn).
[PDF]
Cite:Mang Zhao, Xiangning Chen, Yun Ge, Shuang Yu, and Ying Chen, "A Precision, Large Scale, Anti-Noise Method for Correction of Speech Fundamental Frequency," Journal of Advances in Computer Networks vol. 2, no. 1, pp. 67-70, 2014.