Digital Speech Processing

Explore fundamentals of human speech production, model speech as a system, and examine the vocal tract, speech properties, types, acrostic theory, and acrostic phonetics, with Matlab waveform examples.

Explore how the lungs push air through the vocal cords to produce vibration and pitch, while the nonuniform vocal tract shapes sound via mouth and nasal cavities.

Explore the properties of speech, including silence and the distinction between voiced and unvoiced sounds, and examine their pitch, harmonics, and spectral patterns in time and frequency domains.

Learn the acoustic theory of speech production through a block-diagram model linking an excitation generator and glottal pulse to a vocal tract model, producing quasiperiodic voiced and unvoiced speech.

this lecture examines nasals in time-domain and spectrum, showing how closing the oral track causes cancellations and missing frequencies. it notes most speech content lies below four kilohertz.

Explore the anatomy and function of the ear, from outer to inner, and how we perceive loudness, pitch, and masking, with applications to speech processing.

Model the vocal tract as lossless tubes and a vocal-tract filter, capturing glottal pulse, pitch, and formants. Estimate per-segment parameters from speech for a compact representation.

Explore time-domain speech analysis in 20–30 ms frames, computing short-term energy, short-time magnitude, zero crossings, autocorrelation, and the average magnitude difference function to estimate pitch and voice parameters.

Explore how the Matlab demonstration analyzes speech in the time domain with a rectangular window. See block-wise processing compute squares and magnitudes, then apply a threshold to detect speech segments.

Explore frequency domain methods for speech processing using short-time windowing to build spectrograms and analyze time–frequency content. Compare overlap-add and frequency-bank synthesis, illustrating time and frequency resolution trade-offs.

Demonstrates linear predictive analysis of speech, modeling the vocal tract as a filter and estimating LPC coefficients with autocorrelation and Levinson algorithm to synthesize speech from few samples.

Explore how speech appears in time and frequency, extract LPC-based parameters and energy and magnitude measures, and apply them to speech silence detection, pitch, formant estimation, and basic compression.