1. Isolated Digit Recognition using Discrete HMMs (code). Use the given features also extract your own features and compare the results.
2. Use the HMMs trained in task 1 to recognize continuous digits. You need to concatenate the HMMs trained in task 1 to recognize continuous digits. Use only the given features.
Datasets:
Digit dataset: This dataset consists of spoken utterances. The MFCC feature files and the original .wav files given.
Data: download here, Group Mapping: Download here
Continues digits dataset:
● Download development data from here and test data from here.
● The data contains directories with the group numbers.
● Each directory contains MFCC features from utterances of multiple digits
(corresponding to the isolated digits assigned to your batch).
● The set of digits uttered are given below: symbol - uttered word 1 - one 2 - two 3 - three 4 - four 5 - five 6 - six 7 - seven 8 - eight 9 - nine z - zero o - o
● In development data, the file name represents spoken digits. Eg. In file 534.mfcc, the digits spoken are five three four.
● Test data consists of 5 unlabeled sequences (blind data). Provide the possible sequence of digits obtained in the report.
Feature File Format:
● The data given are the MFCC features of speech audio.
● Structure of MFCC file: The first line of the MFCC file contains two space-separated integers. First integer NC - The dimension of the feature vector (The number of MFC coefficients) Second integer NF - The number of frames, the .wav file is divided into.
● The next NF rows contain the MFCC features of dimension NC. Each row corresponds to a feature vector in the sequence. Please note that NF varies with the example.
