Linear Prediction of Speech
Why Read This Book
You should read this book if you want a rigorous, foundational treatment of linear predictive coding (LPC) as applied to speech: you will learn the mathematical derivations, algorithmic implementations, and practical assumptions behind LPC models. It uniquely combines statistical signal‑processing theory with speech‑specific modeling and algorithms (e.g., Levinson‑Durbin, lattice forms) that remain the basis for modern speech coding and analysis.
Who Will Benefit
Graduate students, researchers, and engineers working on speech processing, speech coding, or spectral analysis who need a deep theoretical and algorithmic grounding in linear prediction.
Level: Advanced — Prerequisites: Undergraduate signals & systems, linear algebra, basic probability/statistics, and familiarity with discrete‑time signal processing (z‑transform, convolution, Fourier analysis).
Key Takeaways
- Derive the LPC model from speech production and statistical estimation principles.
- Implement core LPC algorithms such as the autocorrelation and covariance methods and the Levinson‑Durbin recursion.
- Design and realize LPC-based filter structures including lattice and direct forms for analysis and synthesis.
- Apply spectral analysis techniques based on linear prediction for formant estimation and high‑resolution spectral estimation.
- Evaluate and quantify performance tradeoffs in prediction order, stability, quantization, and coding for speech applications.
Topics Covered
- Introduction: speech production and motivations for linear prediction
- Mathematical formulation of linear prediction and optimality criteria
- Autocorrelation method and properties
- Covariance method and its implications
- Levinson‑Durbin algorithm and numerical issues
- Lattice structures and reflection coefficients
- Spectral analysis and formant estimation using LPC
- Filter realization, stability and implementation considerations
- Quantization, coding and low‑rate speech representation
- Applications: synthesis, coding, and recognition
- Extensions: adaptive prediction and time‑varying models
- Performance evaluation, limitations and practical recommendations
- Appendices: mathematical tools and derivations
Languages, Platforms & Tools
How It Compares
More focused on the theory and algorithmic derivation of LPC than broader texts like Rabiner & Schafer's 'Digital Processing of Speech Signals'; for a concise tutorial perspective, compare with J. Makhoul's review papers on linear prediction.
