QBSH-code-snippets

dtw1d_std.py

# DTW distance
# Change dtwPath
# open terminal, go to your/path/to/sankalp/dtw, then compile: python setup.py build_ext --inplace
# use function dtw1d_std, inputs are two pitch contour lists, use udist as distance

import os,sys

fileDir = os.path.dirname(os.path.realpath('__file__'))
dtwPath = os.path.join(fileDir, 'your/path/to/sankalp/dtw')
dtwPath = os.path.abspath(os.path.realpath(dtwPath))
sys.path.append(dtwPath)

import dtw

def dtw1d_std(x,y):
    """
    1D standard DTW
    input: x, y, two lists
    output: udist, unnormalized distance
    """

    configuration = {}
    configuration['Output'] = 3
    configuration['Ldistance'] = {}
    configuration['Ldistance']['type'] = 0

    udist,_,path = dtw.dtw1d(x,y,configuration)
    return udist,path

evaluation.py

# Evaluation functions

def MRR(list_rank):
    '''
    mean reciprocal rank
    :param list_rank: [rank_0,rank_1,...], each one starting from 1
    :return:
    '''

    return np.sum(1.0/np.array(list_rank))/len(list_rank)

def topXhit(X,list_rank):
    '''
    :param X: if you want to calculate top 5 hit, assign X=5
    :param list_rank: 
    :return: 
    '''
    counter = 0
    for rank in list_rank:
        if rank <= X:
            counter+=1
    return counter/float(len(list_rank))

extract_pitch_pyin.py

# pitch contour extraction
# install pyin https://code.soundsoftware.ac.uk/projects/pyin/files
# pyin is a vamp plug-in, so you need to do http://www.vamp-plugins.org/download.html#install
# to use pyin in python, you need to install https://pypi.python.org/pypi/vamp/1.0.0
# to load audio, you need Essentia

def hz2cents(pitchInHz, tonic=261.626):
    """
    convert hz to cents
    input: an array in hz
    """
    cents = 1200*np.log2(1.0*pitchInHz/tonic)
    return cents

# essentia load audio
loader = ess.MonoLoader(filename=filename_wav, downmix = 'left', sampleRate = sr)
audio = loader()

# extract pitch contour, by a fixed sample rate "sr"
# hopsize = 1024 frames
# framesize = 2048 frames
data = vamp.collect(audio, sr, "pyin:pyin", output='smoothedpitchtrack')
pitch = data['vector'][1]

# convert Hz to cents
pitchInCents = hz2cents(pitch)