Implementation of Pre-Emphasis Analog Filter on Raspberry Pi Using Zero-Order Hold Discretization as a Pre-Processing to Humanoids’ Commands Recognition

Abstract

Intisari – Dalam teknologi yang menggunakan pengolahan sinyal wicara, tapis pre-emphasis sering digunakan sebelum proses ekstraksi ciri. Tapis ini dapat memperbesar amplitudo dari sinyal pada frekuensi tinggi sambil tetap mempertahankan bagian sinyal pada frekuensi rendah. Dalam penelitian ini, kami menerapkan tapis pre-emphasis analog yang dikembangkan dari tapi pelolos frekuensi tinggi. Lalu, filter tersebut didiskretkan menggunakan metode zero-order hold (ZOH). Kedua filter tersebut memiliki pole yang menunjukkan kestabilan. Nilai pole dari tapis diskret selalu di titik nol, sedangkan nilai zero -0.9375. Hasil rekonstruksi dengan menggunakan tapis ­de-emphasis menunjukkan nilai mean-squared error nol yang menyimpulkan bahwa strategi menggunakan metode frame-by-frame untuk implementasi tapis pre-emphasis sangat dianjurkan.

Kata kunci – pre-emphasis, de-emphasis, mean-squared error, tapis, wicara

Abstract – In speech processing technology, the pre-emphasis filter is often used before feature extraction. This filter can emphasize amplitude with high frequencies while preserving the low frequencies side. In this research, we implemented a complete analog pre-emphasis filter from a conventional analog high-pass filter. Then the analog pre-emphasis filter is discretized using zero-order hold (ZOH). The analog and discrete forms have been shown stable according to pole location. The pole of the discrete form is always at the origin, while zero is -0.9375. The reconstruction result, with the help of the de-emphasis filter, shows no discrepancies over the original, hence the mean-squared error is zero which establishes the conclusion that the frame-by-frame method applies to the pre-emphasis filter is lossless and preferable.

Keywords – pre-emphasis, de-emphasis, mean-squared error, filter, speech