GMM-based speaker verification

**Guidelines: **

**GMM-based speaker verification**- Compute the matrix of MFCC for the utterances S0-S9 from the corpus recorded at the first seminar same way as it was done at last seminar, see VQ-based speaker identification.
- Compute GMM models of cepstrum variability for your own voice using the function
(example of its usage is in gmm_example.m and in the previous seminar for vowel classification).**gmdistribution.fit** - Use number of mixtures
*m=4*. - Compute cepstrum of utterances of 2 different speakers (i.e. for your voice and a voice of other speaker). Do not use same utterances as for the training of GMM (i.e. use utterances Z0-Z9 from recorded database).
- Using MATLAB function
**pdf**compute log-likelihoods for all short-time frames of analyzing speakers. - 1st checked results: Display:
- all
**short-time values as well as the mean of the score computed on the basis of emitted logarithmic likelihood for all test utterances for your own voice**, - short-time values as well as the mean of the score
**test utternaces for other speaker**. - the result of
**verification for other speakers**.

- all

**Improved GMM-based speaker verification**- Repeat above realized computation for extended feature vector use 20 MFCC coefficients without c[0].
- 2nd checked result: Display:
- short-time values as well as the mean of the score for your
and other voice in the case when
**extended feature vector of 20 MFCC coefficients**is used.

- short-time values as well as the mean of the score for your
and other voice in the case when
- Extended feature vector by differential features (i.e. numeric estimation of the 1st derivative of the order 2 computed from 2 neighbouring frames). Use the following function diffceps for the estimation of delta features.
- 3rd checked result: Display:
- the verification results using the same procedure realized before for the features vector extended by delta parameters.

**On-line GMM-based speaker verification**- In the case of
**off-line and on-line**verification try to apply**cepstral normalization (CMN)**. To realize cepstral mean subtraction in the utterance, use to following function cmn. - 4th checked result::
- Working on-line speaker verification with possible choice of CMN application using GMM modelling.

- In the case of