06278c94dd2e9d572840ee2133f0d9b366effaf8,pyannote/audio/applications/feature_extraction.py,,extract,#Any#Any#Any#Any#Any#Any#,92

Before Change

    protocol = database.get_protocol(task_name, protocol_name, progress=True)

    if isinstance(protocol, SpeakerDiarizationProtocol):
        items = itertools.chain(protocol.train(),
                                protocol.development(),
                                protocol.test())

After Change

    f.attrs["dimension"] = dimension
    f.close()

    for subset in ["development", "test", "train"]:

        try:
            file_generator = getattr(protocol, subset)()
            first_item = next(file_generator)
        except NotImplementedError as e:
            continue

        file_generator = getattr(protocol, subset)()

        for current_file in file_generator:

            uri = get_unique_identifier(current_file)
            path = Precomputed.get_path(experiment_dir, current_file)

            if os.path.exists(path):
                continue

            try:
                // NOTE current_file contains the "channel" key
                features = feature_extraction(current_file)
            except PyannoteFeatureExtractionError as e:
                if robust:
                    msg = "Feature extraction failed for file "{uri}"."
                    msg = msg.format(uri=uri)
                    warnings.warn(msg)
                    continue
                else:
                    raise e

            if features is None:
                msg = "Feature extraction returned None for file "{uri}"."
                msg = msg.format(uri=uri)
                if not robust:
                    raise PyannoteFeatureExtractionError(msg)
                warnings.warn(msg)
                continue

            data = features.data

            if np.any(np.isnan(data)):
                msg = "Feature extraction returned NaNs for file "{uri}"."
                msg = msg.format(uri=uri)
                if not robust:
                    raise PyannoteFeatureExtractionError(msg)
                warnings.warn(msg)
                continue

            // create parent directory
            mkdir_p(os.path.dirname(path))

            f = h5py.File(path)
            f.attrs["start"] = sliding_window.start
            f.attrs["duration"] = sliding_window.duration
            f.attrs["step"] = sliding_window.step
            f.attrs["dimension"] = dimension
            f.create_dataset("features", data=data)
            f.close()


def main():

    arguments = docopt(__doc__, version="Feature extraction")

Instances