def __init__(self, model_path, output_window_size):
"""
Construct a featurizer from the given ELL model.
The featurizer will ensure it returns an even number of
frames to ensure it files the given output_window_size.
model_path - the path to the ELL compiled model
output_window_size - the classifier window size
"""
self.using_map = False
if os.path.splitext(model_path)[1] == ".ell":
import compute_ell_model as ell
self.model = ell.ComputeModel(model_path)
self.using_map = True
else:
import compiled_ell_model as ell
self.model = ell.CompiledModel(model_path)
self.logfile = None
self.output_window_size = output_window_size
ts = self.model.input_shape
self.input_shape = (ts.rows, ts.columns, ts.channels)
ts = self.model.output_shape
self.output_shape = (ts.rows, ts.columns, ts.channels)
self.input_size = int(self.model.input_shape.Size())
self.output_size = int(self.model.output_shape.Size())
self.frame_count = 0
self.eof = True
self.total_time = 0
def __init__(self,
model_path,
categories_file,
threshold=0,
smoothing_window=0,
ignore_list=[]):
"""
Initialize the new AudioClassifier.
model - the path to the ELL model module to load.
categories_file - the path to a text file containing strings labels for each prediction
threshold - threshold for predictions, (default 0).
smoothing_window - controls the size of the smoothing window (defaults to 0).
ignore_list - list of category labels to ignore (like 'background' or 'silence')
"""
self.smoothing_window = None
if smoothing_window is not None:
self.smoothing_window = float(smoothing_window)
self.threshold = threshold
self.categories = None
if isinstance(ignore_list, str):
self.ignore_list = [x.strip() for x in ignore_list.split(',')]
elif isinstance(ignore_list, list):
self.ignore_list = ignore_list
elif ignore_list is None:
self.ignore_list = []
else:
raise Exception(
"Expecting ignore list to be a comma separated list or a python list of strings"
)
if categories_file:
with open(categories_file, "r") as fp:
self.categories = [e.strip() for e in fp.readlines()]
self.using_map = False
if os.path.splitext(model_path)[1] == ".ell":
import compute_ell_model as ell
self.model = ell.ComputeModel(model_path)
self.using_map = True
else:
import compiled_ell_model as ell
self.model = ell.CompiledModel(model_path)
self.logfile = None
ts = self.model.input_shape
self.input_shape = (ts.rows, ts.columns, ts.channels)
ts = self.model.output_shape
self.output_shape = (ts.rows, ts.columns, ts.channels)
self.input_size = int(self.model.input_shape.Size())
self.output_size = int(self.model.output_shape.Size())
self.smoothing_items = []
self.total_time = 0
self.count = 0
def __init__(self, model):
if isinstance(model, str):
if os.path.splitext(model)[1] == ".ell":
import compute_ell_model as ell
self.model = ell.ComputeModel(model)
self.using_map = True
else:
import compiled_ell_model as ell
self.model = ell.CompiledModel(model)
else:
import compute_ell_model as ell
self.model = ell.ComputeModel(model)
self.using_map = True
def __init__(self,
model_path,
categories_file,
ignore_list=[],
threshold=0,
smoothing_delay=0.2):
"""
Initialize the new AudioClassifier.
model - the path to the ELL model module to load.
categories_file - the path to a text file containing strings labels for each prediction
ignore_list - a list of prediction indices to ignore
threshold - threshold for predictions, (default 0).
smoothing_delay - controls the size of this window (defaults to 0.2 seconds).
"""
self.smoothing_delay = smoothing_delay
self.ignore_list = ignore_list
if not self.ignore_list:
self.ignore_list = []
self.threshold = threshold
self.categories = None
if categories_file:
with open(categories_file, "r") as fp:
self.categories = [e.strip() for e in fp.readlines()]
self.using_map = False
if os.path.splitext(model_path)[1] == ".ell":
import compute_ell_model as ell
self.model = ell.ComputeModel(model_path)
self.using_map = True
else:
import compiled_ell_model as ell
self.model = ell.CompiledModel(model_path)
self.logfile = None
ts = self.model.input_shape
self.input_shape = (ts.rows, ts.columns, ts.channels)
ts = self.model.output_shape
self.output_shape = (ts.rows, ts.columns, ts.channels)
self.input_size = int(self.model.input_shape.Size())
self.output_size = int(self.model.output_shape.Size())
self.items = []
self.start_time = None
self.total_time = 0
self.count = 0
def __init__(self, model_path, output_window_size):
"""
Construct a featurizer from the given ELL model.
The featurizer will ensure it returns an even number of
frames to ensure it files the given output_window_size.
model_path - the path to the ELL compiled model
output_window_size - the classifier window size
"""
self.using_map = False
if os.path.splitext(model_path)[1] == ".ell":
import compute_ell_model as ell
self.model = ell.ComputeModel(model_path)
self.using_map = True
else:
import compiled_ell_model as ell
self.model = ell.CompiledModel(model_path)
self.logfile = None
self.output_window_size = output_window_size
ts = self.model.input_shape
self.input_shape = (ts.rows, ts.columns, ts.channels)
ts = self.model.output_shape
self.output_shape = (ts.rows, ts.columns, ts.channels)
self.input_size = int(self.model.input_shape.Size())
self.output_size = int(self.model.output_shape.Size())
# if the featurizer is buffering input internally, then we need
# to "prime" the input to that buffer until the buffer is full
# so we don't return featurized "zeros" because the buffer not yet full.
self.drop_frames = 0
value = self.get_metadata("window_size")
if value:
buffer_size = int(value)
self.drop_frames = int(buffer_size / self.input_size)
self.dropped = 0
self.reset()