IIR FILTER ARCHITECTURES FOR NOISY NARROWBAND SPEECH ENHANCEMENT: COMPARING BUTTERWORTH, CHEBYSHEV TYPE I, AND ELLIPTIC DESIGNS

Narrowband speech signals occupying the 300–3400 Hz telephony band are routinely degraded by 50/60 Hz power-supply hum, additive white Gaussian noise (AWGN), and high-frequency microphone hiss. While finite impulse response (FIR) filters are preferred for ECG processing due to their linear-phase property, infinite impulse response (IIR) filters offer substantially lower computational order for equivalent stopband attenuation — a critical advantage in real-time voice processing on resource-constrained devices such as VoIP gateways, hearing aids, and mobile telephony chips. This paper presents a MATLAB-based comparative study of three IIR bandpass filter designs — Butterworth (N = 6), Chebyshev Type I (N = 5), and Elliptic/Cauer (N = 4) — evaluated against a synthetic voiced speech signal contaminated with multi-source noise at SNR in = 8 dB. Performance is assessed via output SNR, ∆SNR, MSE, and group delay variation. Spectrogram analysis and pole-zero diagrams complement the quantitative metrics. Results show that the Elliptic filter achieves the highest ∆SNR (11.7 dB) with the lowest pole count (8) and sharpest transition rolloff, at the cost of nonlinear group delay that must be managed in perceptually sensitive applications. Complete MATLAB source code is provided for full reproducibility.

Перевод пока недоступен