CS-578

Voice Processing

Postgraduate

Discrete Time Signal Processing Framework
Production and Classification of Speech Sounds
Acoustics of Speech Production
Linear Prediction: analysis, synthesis, and modifications of speech: Autocorrelation method, Covariance method
Glottal signal models, Inverse filtering
Sinusoidal Models: analysis, synthesis, and modifications of speech
Sinusoidal Model, Harmonic Model, Adaptive Sinusoidal Models
Speech Transformations. Time scaling, Pitch scaling
Speech Coding, Coding schemes: Scalar quantization, Vector quantization
Speech Enhancement: Spectral Subtraction, Wiener Filter
Speech Intelligibility Enhancement: Spectral Shaping, Dynamic Range Compression
Speaker Identification: EM algorithm, Gaussian Mixture Models
Speech Synthesis: Neural Networks, Hidden Markov Models, Linear Dynamical Models
Machine Learning techniques: Deep Learning
 
Monthly MATLAB/Python projects on speech processing are mandatory.

LEARNING OUTCOMES
1. Knowledge: Having attended and succeeded in the course, the student is able to recognize the characteristics of human speech, the types of its phonemes, do research and implement speech processing models in time and frequency domain, and produce techniques for speech analysis, processing, and synthesis.
2. Understanding: Having attended and succeeded in the course, the student has understood the most common methods of speech signal analysis, its general characteristics, and its behavior in time and frequency domain.
3. Application: Having attended and succeeded in the course, the student is able to implement systems for speech analysis, modification, and synthesis, speaker recognition, speech enhancement, speech intelligibility enhancement, pathological speech recognition, sinusoidal modeling and speech modification, and speech coding.
4. Analysis: Having attended and succeeded in the course, the student is able to analyze the speech signal into components, use the corresponding processing techniques, and extract speech features for the aforementioned applications.
5. Synthesis: Having attended and succeeded in the course, the student is able to implement new speech processing algorithms, adapted to required resources, performance metrics, and accuracy requirements of the respective input data and the respective application and combine them with existing ones.
6. Evaluation: Having attended and succeeded in the course, the student is able to measure/quantify the quality of solutions to speech processing problems and compare these solutions to other existing ones.

STUDENT PERFORMANCE EVALUATION
Class attendance/participation
Exercises/projects
Oral/written exams
 

6

CS-370

http://www.csd.uoc.gr/~hy578