Georgi Dzhambazov

To eradicate the error about downsample change ``` from theano.tensor.signal import downsample ... pooled_out = downsample.max_pool_2d( ... ) ``` to ``` from theano.tensor.signal import pool ... pooled_out = pool.pool_2d( ......

Before starting, check the [readme](https://github.com/georgid/AlignmentDuration/blob/production_no_melody/src/align/readme.design.classes#L64) until the Lyrics class I. Reimplement essential methods: 1. Reimplement [creation of a transition probability matrix](https://github.com/georgid/AlignmentDuration/blob/production_no_melody/src/hmm/continuous/_HMM.py#L240). It needs as parameter the `statesNetwork` which is a...

Step 3. done in the [fork](https://github.com/georgid/essentia/commit/1b52dc58abf0c2592da2312cee58246998623216 )

pyflakes done

One way is to do a static library out of essentia using .waf and the flag --build-static. In order not to compile the whole essentia, one can include or exclude...

emphasis reproduced [here](https://github.com/georgid/AlignmentDuration/commit/c381aa066708a05072daf2d9eb34f87e4994a9d5) , but still there is an error in the differernce between mfccs. The difference can be seen by running [this code](https://github.com/georgid/mfcc-htk-an-librosa/blob/master/mfcc-htk-many-parameters.py) Seems for acapella the difference leads...

the size of fft should be the same as the one in htk, so when reducing the sampling rate, it whould be taken into account.

remove align._PhonemeBase.PhonemeBase.getNumStates and replace by return 1 move this logic ``` ``` stateIndices = range( phoneme.getNumStates()) if ParametersAlgo.ONLY_MIDDLE_STATE and len(phoneme.model.states) == 3: # take only middle state. 1 is middle...

put in method hmm.continuous.GMHMM.GMHMM._set_GMMs. could be optimized more: eliminate weights, models, and covars

the problem is that we cannot install python as dependency or that the setup.py is not correct? I assume the second: Cython is necessary dependency of https://github.com/mtg/sms-tools/ which I use....