Publication detail

Efficient spectral power estimation on an arbitrary frequency scale

ZÁPLATA, F. KASAL, M.

Original Title

Efficient spectral power estimation on an arbitrary frequency scale

English Title

Efficient spectral power estimation on an arbitrary frequency scale

Type

journal article in Web of Science

Language

en

Original Abstract

The Fast Fourier Transform is a very efficient algorithm for the Fourier spectrum estimation, but has the limitation of a linear frequency scale spectrum, which may not be suitable for every system. For example, audio and speech analysis needs a logarithmic frequency scale due to the characteristic of a humans ear. The Fast Fourier Transform algorithms are not able to efficiently give the desired results and modified techniques have to be used in this case. In the following text a simple technique using the Goertzel algorithm allowing the evaluation of the power spectra on an arbitrary frequency scale will be introduced. Due to its simplicity the algorithm suffers from imperfections which will be discussed and partially solved in this paper. The implementation into real systems and the impact of quantization errors appeared to be critical and have to be dealt with in special cases. The simple method dealing with the quantization error will also be introduced. Finally, the proposed method will be compared to other methods based on its computational demands and its potential speed.

English abstract

The Fast Fourier Transform is a very efficient algorithm for the Fourier spectrum estimation, but has the limitation of a linear frequency scale spectrum, which may not be suitable for every system. For example, audio and speech analysis needs a logarithmic frequency scale due to the characteristic of a humans ear. The Fast Fourier Transform algorithms are not able to efficiently give the desired results and modified techniques have to be used in this case. In the following text a simple technique using the Goertzel algorithm allowing the evaluation of the power spectra on an arbitrary frequency scale will be introduced. Due to its simplicity the algorithm suffers from imperfections which will be discussed and partially solved in this paper. The implementation into real systems and the impact of quantization errors appeared to be critical and have to be dealt with in special cases. The simple method dealing with the quantization error will also be introduced. Finally, the proposed method will be compared to other methods based on its computational demands and its potential speed.

Keywords

Goertzel algorithm, Mel frequency cepstral coefficients, MFCCs, Q-constant transform, Spectral power estimation.

RIV year

2015

Released

01.04.2015

Pages from

178

Pages to

184

Pages count

7

BibTex


@article{BUT111234,
  author="Filip {Záplata} and Miroslav {Kasal}",
  title="Efficient spectral power estimation on an arbitrary frequency scale",
  annote="The Fast Fourier Transform is a very efficient algorithm for the Fourier spectrum estimation, but has the limitation of a linear frequency scale spectrum, which may not be suitable for every system. For example, audio and speech analysis needs a logarithmic frequency scale due to the characteristic of a humans ear. The Fast Fourier Transform algorithms are not able to efficiently give the desired results and modified techniques have to be used in this case. In the following text a simple technique using the Goertzel algorithm allowing the evaluation of the power spectra on an arbitrary frequency scale will be introduced. Due to its simplicity the algorithm suffers from imperfections which will be discussed and partially solved in this paper. The implementation into real systems and the impact of quantization errors appeared to be critical and have to be dealt with in special cases. The simple method dealing with the quantization error will also be introduced. Finally, the proposed method will be compared to other methods based on its computational demands and its potential speed.",
  chapter="111234",
  doi="10.13164/re.2015.0178",
  howpublished="print",
  number="1",
  volume="2015",
  year="2015",
  month="april",
  pages="178--184",
  type="journal article in Web of Science"
}