Pruning Algorithms for Multilayer Perceptrons Tailored for Speech/Non-Speech Classification in Digital Hearing Aids

Álvarez L, Alexandre E, Rosa-Zurera M. Pruning Algorithms for Multilayer Perceptrons Tailored for Speech/Non-Speech Classification in Digital Hearing Aids. In: AES Convention 126. Audio Engineering Society; 2009. Paper 7736. Available from: https://aes.org/publications/elibrary-page/?id=14932

@inproceedings{Alvarez2009_14932,
  author = {Álvarez, Lorena and Alexandre, Enrique and Rosa-Zurera, Manuel},
  title = {{Pruning Algorithms for Multilayer Perceptrons Tailored for Speech/Non-Speech Classification in Digital Hearing Aids}},
  booktitle = {AES Convention 126},
  note = {Paper 7736},
  year = {2009},
  month = may,
  publisher = {Audio Engineering Society},
  url = {https://aes.org/publications/elibrary-page/?id=14932}
}

TY  - CPAPER
TI  - Pruning Algorithms for Multilayer Perceptrons Tailored for Speech/Non-Speech Classification in Digital Hearing Aids
AU  - Álvarez, Lorena
AU  - Alexandre, Enrique
AU  - Rosa-Zurera, Manuel
T2  - AES Convention 126
M1  - Paper 7736
PY  - 2009
DA  - 2009/05/06
UR  - https://aes.org/publications/elibrary-page/?id=14932
PB  - Audio Engineering Society
LA  - en
AB  - This paper explores the feasibility of using different pruning algorithms for multilayer perceptrons (MLPs) applied to the problem of speech/non-speech classification in digital hearing aids. A classifier based on MLPs is considered the best option in spite of its presumably high computational cost. Nevertheless, its implementation has been proven to be feasible: it requires some trade-offs involving a balance between reducing the computational demands (that is, the number of neurons) and the quality perceived by the user. In this respect, this paper will focus on the design of three novel pruning algorithms for MLPs, which attempt to converge to the minimum complexity network (that is, the lowest number of neurons in the hidden layer) without degrading the performance of it. The results obtained with the proposed algorithms will be compared with those obtained when using another pruning algorithm proposed in the literature.
ER  -