Ebook: Bandwidth Extension of Speech Signals

Bandwidth Extension of Speech Signals provides discussion on different approaches for efficient and robust bandwidth extension of speech signals while acknowledging the influence of noise corrupted real-world signals. The book describes the theory and methods for quality enhancement of clean speech signals and distorted speech signals such as those that have undergone a band limitation, for instance, in a telephone network. Problems and the respective solutions are discussed in regards to different approaches. The different approaches are evaluated and robustness issues for a real-time implementation are covered as well. The book includes topics related to speech coding, pattern- / speech recognition, speech enhancement, statistics and digital signal processing in general.

---

This book describes the study and development of several methods for the bandwidth extension of speech signals. The proposed methods include the extraction of speech-model based parameters like cepstral coefficients, auto-regressive parameter (all-pole-filter coefficients), line spectral pairs (LSP) or mel-frequency cepstral coefficients (MFCC). These parameters and their properties are briefly presented. Besides these parametric representations of the spectral envelope some other scalar speech features like the zero-crossing rate, gradient index, pitch analysis, local kurosis, spectral centroid and energy based features are introduced.The book also focuses on the performance of these methods, using the objective spectral distortion measures. A significance analysis is also drawn forthe subjective evaluation as well as a rank correlation between the subjective results and the respective objective distortion measures to evaluate the significance of these objective measures. Finally, the most promising method is implemented in a real-time environment using appropriate hardware.