HMM Code for MATLAB Speech Recognition

dtw - DTW algorithm demonstration program showcases the Dynamic Time Warping algorithm implementation in MATLAB, demonstrating its application in speech recognition through path visualization and distance calculation between time series. mfcc.m - MFCC parameter calculation program computes Mel Frequency Cepstral Coefficients using frame-based processing, FFT transformation, Mel-filterbank application, and DCT transformation to extract speech features for recognition systems. dtw.m - Basic DTW algorithm implements the fundamental dynamic programming approach to find the optimal alignment path between two time sequences, calculating minimum cumulative distance through matrix operations. dtw2.m - Optimized DTW algorithm enhances the basic implementation with computational optimizations like bandwidth limiting and pruning techniques to improve efficiency while maintaining alignment accuracy. testdtw.m - DTW algorithm testing program provides comprehensive testing framework with multiple test cases, performance metrics calculation, and visualization tools to validate DTW algorithm performance across different datasets. vad.m - Endpoint detection program identifies speech segment boundaries using energy-based thresholds and zero-crossing rate analysis, crucial for preprocessing in speech recognition pipelines. cdhmm - Continuous Gaussian Mixture HMM demonstration program illustrates the implementation of continuous density HMMs with Gaussian mixture models for state output distributions, including training and decoding processes. pdf.m - Gaussian probability density function computes multivariate Gaussian probabilities using covariance matrices and mean vectors, fundamental for statistical modeling in HMMs. mixture.m - Gaussian mixture output probability calculates the likelihood of observation sequences under mixture models by combining weighted Gaussian components through logarithmic sum operations. inithmm.m - HMM parameter initialization sets up initial state probabilities, transition matrices, and emission parameters using uniform distributions or data-driven approaches for model training. getparam.m - Observation sequence parameter calculation extracts statistical features and prepares observation vectors for HMM training, including normalization and dimension handling. viterbi.m - Viterbi algorithm for speech recognition implements the dynamic programming solution for finding the most likely state sequence through trellis computation and path backtracking. baum.m - Baum-Welch training algorithm (single iteration) performs one EM algorithm iteration for HMM parameter re-estimation, updating transition and emission probabilities using forward-backward probabilities. main.m - Multiple HMM training main program coordinates the training of multiple Hidden Markov Models simultaneously, managing data partitioning and model synchronization for complex recognition tasks. train.m - Single HMM training program optimizes parameters for individual HMMs using iterative Baum-Welch algorithm with convergence checking and parameter smoothing techniques. recog.m - Recognition program performs speech classification by computing likelihood scores against trained HMM models using Viterbi decoding or forward algorithm for pattern matching. vad.m - Endpoint detection program (repeated entry) implements voice activity detection with frame-based analysis and decision logic for robust speech boundary identification. mfcc.m - MFCC parameter calculation program (repeated entry) extracts frequency-domain features through spectral analysis and cepstral transformation for speech representation. samples.mat - Chinese digit 0-9 recordings contains audio samples of Mandarin digits recorded by the author, providing test data for speech recognition algorithm development and validation. hmm.mat - HMM training results stores trained model parameters including state transition probabilities, emission distributions, and mixture weights from a complete training session. record - Auxiliary recording program provides audio recording interface with real-time monitoring and file management capabilities for data collection. record.m - Recording script file controls audio recording parameters including sample rate, duration, and file storage paths through MATLAB's audio input functions. record.fig - Recording program GUI offers graphical interface with control buttons and visual feedback for intuitive audio recording operations. sample.m - Recording callback function handles event-driven operations during recording, such as real-time waveform display and processing triggers.