base.py 文件源码

def predict(self, filename=None, y=None, sr=None, outputs=None):
        '''Predict annotations

        Parameters
        ----------
        filename : str (optional)
            Path to audio file

        y, sr : (optional)
            Audio buffer and sample rate

        outputs : (optional)
            Pre-computed model outputs as produced by `CremaModel.outputs`.
            If provided, then predictions are derived from these instead of
            `filename` or `(y, sr)`.


        .. note:: At least one of `filename`, `y, sr` must be provided.

        Returns
        -------
        jams.Annotation
            The predicted annotation
        '''

        # Pump the input features
        output_key = self.model.output_names[0]

        if outputs is None:
            outputs = self.outputs(filename=filename, y=y, sr=sr)

        # Invert the prediction.  This is always the first output layer.
        ann = self.pump[output_key].inverse(outputs[output_key])

        # Populate the metadata
        ann.annotation_metadata.version = self.version
        ann.annotation_metadata.annotation_tools = 'CREMA {}'.format(version)
        ann.annotation_metadata.data_source = 'program'
        ann.duration = librosa.get_duration(y=y, sr=sr, filename=filename)

        return ann