a6f283c27fdefcbf4b3fc844206cb698786d7eee,nussl/separation/nmf_mfcc.py,NMF_MFCC,__init__,#NMF_MFCC#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#Any#,52

Before Change

            raise ValueError("mfcc_range is not set correctly! Must be a tuple or list with min and max, or int (max)")

        // If provided, add random_seed to kmeans_kwargs
        if self.random_seed is not None:
            if self.kmeans_kwargs is None:
                self.kmeans_kwargs = {"random_state": self.random_seed}
            if "random_state" not in self.kmeans_kwargs or self.kmeans_kwargs["random_state"] is None:
                self.kmeans_kwargs["random_state"] = self.random_seed

        // Set number of clusters to number of sources
        if self.kmeans_kwargs is None:
            self.kmeans_kwargs = {"n_clusters": self.num_sources}
        else:
            self.kmeans_kwargs["n_clusters"] = self.num_sources

After Change

            masks (np.array): A Numpy array containing the lists of Binary Mask objects for each channel.

        
    def __init__(self, input_audio_signal, num_sources, num_templates=50, distance_measure="euclidean",
                 num_iterations=50, random_seed=None, kmeans_kwargs=None, convert_to_mono=False,
                 mask_type=mask_separation_base.MaskSeparationBase.BINARY_MASK, mfcc_range=(1, 14), n_mfcc=20):
        super(NMF_MFCC, self).__init__(input_audio_signal=input_audio_signal, mask_type=mask_type)

        self.num_sources = num_sources
        self.num_templates = num_templates
        self.distance_measure = distance_measure
        self.num_iterations = num_iterations
        self.random_seed = random_seed
        self.kmeans_kwargs = kmeans_kwargs
        self.convert_to_mono = convert_to_mono
        self.n_mfcc = n_mfcc
        self.mask_type = mask_type

        self.signal_stft = None
        self.input_audio_signal = input_audio_signal
        self.labeled_templates = None
        self.sources = []
        self.masks = []

        // Convert the stereo signal to mono if indicated
        if self.convert_to_mono:
            self.input_audio_signal.to_mono(overwrite=True, remove_channels=False)

        // Set the MFCC range
        if isinstance(mfcc_range, int) and mfcc_range < n_mfcc:
            self.mfcc_start, self.mfcc_end = 1, mfcc_range
        elif isinstance(mfcc_range, (tuple, list)) and len(mfcc_range) == 2:
            self.mfcc_start, self.mfcc_end = mfcc_range[0], mfcc_range[1]
        else:
            raise ValueError("mfcc_range is not set correctly! Must be a tuple or list with min and max, or int (max)")

        // If kmeans_kwargs does not include the "random_state", use the random_seed instead. Else, use the value
        // provided in the dictionary. If kmeans_kwargs is None, use the random_seed.

        self.kmeans_random_seed = kmeans_kwargs.pop("random_state", random_seed) \
            if isinstance(kmeans_kwargs, dict) else random_seed

        // Initialize the K Means clusterer
        if isinstance(self.kmeans_kwargs, dict):
            self.clusterer = sklearn.cluster.KMeans(n_clusters=self.num_sources, random_state=self.kmeans_random_seed,
                                                    **kmeans_kwargs)
        else:
            self.clusterer = sklearn.cluster.KMeans(n_clusters=self.num_sources, random_state=self.kmeans_random_seed)

    def run(self):
         This function calls TransformerNMF on the magnitude spectrogram of each channel in the input audio signal.
        The templates and activation matrices returned are clustered using K-Means clustering. These clusters are used

Instances