Exemplo n.º 1
0
    def __init__(self, file_struct, feat_type, sr=config.sample_rate,
                 hop_length=config.hop_size, mode=config.cqt.mode,
                 norm=config.cqt.norm, epsilon=config.cqt.epsilon,
                 ref_power=config.cqt.ref_power):
        """Constructor of the class.

        Parameters
        ----------
        file_struct: `msaf.input_output.FileStruct`
            Object containing the file paths from where to extract/read
            the features.
        feat_type: `FeatureTypes`
            Enum containing the type of features.
        sr: int > 0
            Sampling rate for the analysis.
        hop_length: int > 0
            Hop size in frames for the analysis.
        mode: {'stack', 'merge'}
            Mode for flattening multi-track piano-rolls.
        norm: bool
            True to normalize the piano-rolls. False to do nothing.
        epsilon: float
            A small number added to the results to avoid divide by zero errors.
        ref_power: function
            The reference power for logarithmic scaling.
        """
        # Init the parent
        super().__init__(file_struct=file_struct, sr=sr, hop_length=hop_length,
                         feat_type=feat_type)
        # Init the CQT parameters
        if mode not in ('stack', 'merge'):
            raise FeatureParamsError("Wrong value for mode")
        self.mode = mode
        self.norm = norm
        self.epsilon = epsilon
        if ref_power == "max":
            self.ref_power = np.max
        elif ref_power == "min":
            self.ref_power = np.min
        elif ref_power == "median":
            self.ref_power = np.median
        else:
            raise FeatureParamsError("Wrong value for ref_power")
Exemplo n.º 2
0
    def __init__(self,
                 file_struct,
                 feat_type,
                 sr=config.sample_rate,
                 hop_length=config.hop_size,
                 n_fft=config.n_fft,
                 n_mels=config.mfcc.n_mels,
                 n_mfcc=config.mfcc.n_mfcc,
                 ref_power=config.mfcc.ref_power,
                 n_bins=config.cqt.bins,
                 norm=config.cqt.norm,
                 filter_scale=config.cqt.filter_scale):
        """Constructor of the class.

        Parameters
        ----------
        file_struct: `msaf.input_output.FileStruct`
            Object containing the file paths from where to extract/read
            the features.
        feat_type: `FeatureTypes`
            Enum containing the type of features.
        sr: int > 0
            Sampling rate for the analysis.
        hop_length: int > 0
            Hop size in frames for the analysis.
        n_fft: int > 0
            Number of frames for the FFT.
        n_mels: int > 0
            Number of mel filters.
        n_mfcc: int > 0
            Number of mel coefficients.
        ref_power: function
            The reference power for logarithmic scaling.
        """
        # Init the parent
        super().__init__(file_struct=file_struct,
                         sr=sr,
                         hop_length=hop_length,
                         feat_type=feat_type)
        # Init the MFCC parameters
        self.n_fft = n_fft
        self.n_mels = n_mels
        self.n_mfcc = n_mfcc
        # Init the CQT parameters
        self.n_bins = n_bins
        self.norm = norm
        self.filter_scale = filter_scale
        if ref_power == "max":
            self.ref_power = np.max
        elif ref_power == "min":
            self.ref_power = np.min
        elif ref_power == "median":
            self.ref_power = np.median
        else:
            raise FeatureParamsError("Wrong value for ref_power")
Exemplo n.º 3
0
    def __init__(self,
                 file_struct,
                 feat_type,
                 sr=config.sample_rate,
                 hop_length=config.hop_size,
                 n_bins=config.cqt.bins,
                 norm=config.cqt.norm,
                 filter_scale=config.cqt.filter_scale,
                 ref_power=config.cqt.ref_power):
        """Constructor of the class.

        Parameters
        ----------
        file_struct: `msaf.input_output.FileStruct`
            Object containing the file paths from where to extract/read
            the features.
        feat_type: `FeatureTypes`
            Enum containing the type of features.
        sr: int > 0
            Sampling rate for the analysis.
        hop_length: int > 0
            Hop size in frames for the analysis.
        n_bins: int > 0
            Number of frequency bins for the CQT.
        norm: float
            Type of norm to use for basis function normalization.
        filter_scale: float
            The scale of the filter for the CQT.
        ref_power: str
            The reference power for logarithmic scaling.
            See `configdefaults.py` for the possible values.
        """
        # Init the parent
        super().__init__(file_struct=file_struct,
                         sr=sr,
                         hop_length=hop_length,
                         feat_type=feat_type)
        # Init the CQT parameters
        self.n_bins = n_bins
        self.norm = norm
        self.filter_scale = filter_scale
        if ref_power == "max":
            self.ref_power = np.max
        elif ref_power == "min":
            self.ref_power = np.min
        elif ref_power == "median":
            self.ref_power = np.median
        else:
            raise FeatureParamsError("Wrong value for ref_power")
Exemplo n.º 4
0
    def read_features(self, tol=1e-3):
        """Reads the features from a file and stores them in the current
        object.

        Parameters
        ----------
        tol: float
            Tolerance level to detect duration of audio.
        """
        try:
            # Read JSON file
            with open(self.file_struct.features_file) as f:
                feats = json.load(f)

            # Store duration
            if self.dur is None:
                self.dur = float(feats["globals"]["dur"])

            # Check that we have the correct global parameters
            assert (np.isclose(self.dur,
                               float(feats["globals"]["dur"]),
                               rtol=tol))
            assert (self.sr == int(feats["globals"]["sample_rate"]))
            assert (self.hop_length == int(feats["globals"]["hop_length"]))
            assert (os.path.basename(
                self.file_struct.audio_file) == os.path.basename(
                    feats["globals"]["audio_file"]))

            # Check for specific features params
            feat_params_err = FeatureParamsError(
                "Couldn't find features for %s id in file %s" %
                (self.get_id(), self.file_struct.features_file))
            if self.get_id() not in feats.keys():
                raise feat_params_err
            for param_name in self.get_param_names():
                value = getattr(self, param_name)
                if hasattr(value, '__call__'):
                    # Special case of functions
                    if value.__name__ != \
                            feats[self.get_id()]["params"][param_name]:
                        raise feat_params_err
                else:
                    if str(value) != \
                            feats[self.get_id()]["params"][param_name]:
                        raise feat_params_err

            # Store actual features
            self._est_beats_times = np.array(feats["est_beats"])
            self._est_beatsync_times = np.array(feats["est_beatsync_times"])
            self._est_beats_frames = librosa.core.time_to_frames(
                self._est_beats_times, sr=self.sr, hop_length=self.hop_length)
            self._framesync_features = \
                np.array(feats[self.get_id()]["framesync"])
            self._est_beatsync_features = \
                np.array(feats[self.get_id()]["est_beatsync"])

            # Read annotated beats if available
            if "ann_beats" in feats.keys():
                self._ann_beats_times = np.array(feats["ann_beats"])
                self._ann_beatsync_times = np.array(
                    feats["ann_beatsync_times"])
                self._ann_beats_frames = librosa.core.time_to_frames(
                    self._ann_beats_times,
                    sr=self.sr,
                    hop_length=self.hop_length)
                self._ann_beatsync_features = \
                    np.array(feats[self.get_id()]["ann_beatsync"])
        except KeyError:
            raise WrongFeaturesFormatError(
                "The features file %s is not correctly formatted" %
                self.file_struct.features_file)
        except AssertionError:
            raise FeaturesNotFound(
                "The features for the given parameters were not found in "
                "features file %s" % self.file_struct.features_file)
        except IOError:
            raise NoFeaturesFileError("Could not find features file %s",
                                      self.file_struct.features_file)