- All Volumes
- Volume 33, issue 1, 2023 (5 papers)
- Volume 32, Issue 1, 2022 (6 papers)
- Volume 32, Issue 2, 2022 (5 papers)
- Volume 31, Issue 1, 2021 (6 papers)
- Volume 31, Issue 2, 2021 (6 papers)
- Volume 30, Issue 1, 2020 (7 papers)
- Volume 30, Issue 2, 2020 (6 papers)
- Volume 29, Issue 1, 2019 (6 papers)
- Volume 29, Issue 2, 2019 (6 papers)
- Volume 21, Issue 1, 2014 (12 papers)
- Volume 16, Issue 1, 2012 (10 papers)
- Volume 19, Issue 2, 2012 (13 papers)
- Volume 20, Issue 3, 2012 (13 papers)
- Volume 14, Issue 2, 2011 (10 papers)
- Volume 12, Issue1, 2010 (16 papers)
Views : 74 Downloads : 76 Download PDF
Group Delay Moment of Cepstrum for Formant Estimation of High-Pitched Noisy Speech
Corresponding Author : Husne Ara Chowdhury (husna-cse@sust.edu)
Keywords : Deconvolution, Group delay, Spectral root cepstrum, Stabilization, Noise
Abstract :
The estimation task of formant frequencies is challenging for some spectral estimation issues. However, it is significant in the female or child speech-processing arena. This paper proposes a method for calculating the formant frequencies of high-pitched speech employing third-order group delay moment (GDM) of cepstrum. The GDM is a time domain equivalent signal estimated using the inverse discrete Fourier transform (DFT) of group delay spectrum (GDS). The GDS is calculated from the cepstrum. The stabilized spectral root cepstrum (SRC) is used in place of log-based cepstrum to obtain better control of the noisy speech spectrum. The resultant GDM becomes a vocal tract-dominated signal with noise-robust as well. The efficiency of the proposed method has been shown by calculating the formant values of some synthetic vowels against different fundamental frequency variations from 100 Hz to 400 Hz. Additionally, standard F2–F1 plots obtained from the natural vowel sounds of male and female speakers are demonstrated. An utterance from the TIMIT corpus has been utilized to plot the formant contours on the respective spectrogram. The results are likened to two related sophisticated methods. The proposed technique outperforms both approaches, especially when high-pitched speaking in the presence of ambient noise
Published on July 1st, 2023 in Volume 33, issue 1, Applied Sciences and Technology