8th European Conference on Speech Communication and Technology

Geneva, Switzerland
September 1-4, 2003


Performance Evaluation of IFAS-Based Fundamental Frequency Estimator in Noisy Environment

Dhany Arifianto, Takao Kobayashi

Tokyo Institute of Technology, Japan

In this paper, instantaneous frequency amplitude spectrum (IFAS)-based fundamental frequency estimator is evaluated with speech signal corrupted by additive white gaussian noise. A key idea of the IFAS-based estimator is the use of degree of regularity of periodicity in spectrum of speech signal, de- fined by a quantity called harmonicity measure, for band selection in the fundamental frequency estimation. Several frequency band and window length selection methods based on harmonicity measure are assessed to find out better performance. It is shown that the performance of the IFAS-based estimator is maintained at constant error rate about 1% from clean speech data up to 15 dB and about 11% at 0 dB SNR. For both female and male speakers, the IFAS-based estimator outperforms several well-known methods particularly at 0 dB SNR.

Full Paper

Bibliographic reference.  Arifianto, Dhany / Kobayashi, Takao (2003): "Performance evaluation of IFAS-based fundamental frequency estimator in noisy environment", In EUROSPEECH-2003, 2877-2880.