Implements the Linear and Mel Frequency Cepstal Coefficients (MFCC and LFCC)
Algorithms have at least one input and one output. All algorithm endpoints are organized in groups. Groups are used by the platform to indicate which inputs and outputs are synchronized together. The first group is automatically synchronized with the channel defined by the block in which the algorithm is deployed.
| Endpoint Name | Data Format | Nature |
|---|---|---|
| speech | system/array_1d_floats/1 | Input |
| vad | system/array_1d_integers/1 | Input |
| features | system/array_2d_floats/1 | Output |
Parameters allow users to change the configuration of an algorithm when scheduling an experiment
| Name | Description | Type | Default | Range/Choices |
|---|---|---|---|---|
| rate | float64 | 16000.0 |
xxxxxxxxxximport bobimport numpydef normalize_std_array(vector): """Applies a unit mean and variance normalization to an arrayset""" # Initializes variables length = 1 n_samples = len(vector) mean = numpy.ndarray((length,), 'float64') std = numpy.ndarray((length,), 'float64') mean.fill(0) std.fill(0) # Computes mean and variance for array in vector: x = array.astype('float64') mean += x std += (x ** 2) mean /= n_samples std /= n_samples std -= (mean ** 2) std = std ** 0.5 arrayset = numpy.ndarray(shape=(n_samples,mean.shape[0]), dtype=numpy.float64) for i in range (0, n_samples): arrayset[i,:] = (vector[i]-mean) / std return arraysetclass Algorithm: def __init__(self): self.win_length_ms = 20 self.win_shift_ms = 10 self.n_filters = 24 self.n_ceps = 19 self.f_min = 0 self.f_max = 4000 self.delta_win = 2 self.pre_emphasis_coef = 0.95 self.dct_norm = False self.mel_scale = True self.withEnergy = True self.withDelta = True self.withDeltaDelta = True #TODO: find a way to compute this automatically self.rate = 16000 self.features_mask = numpy.arange(0,60) self.normalizeFeatures = True def setup(self, parameters): self.rate = parameters.get('rate', self.rate) wl = self.win_length_ms ws = self.win_shift_ms nf = self.n_filters nc = self.n_ceps f_min = self.f_min f_max = self.f_max dw = self.delta_win pre = self.pre_emphasis_coef rate = self.rate normalizeFeatures = self.normalizeFeatures self.extractor = bob.ap.Ceps(rate, wl, ws, nf, nc, f_min, f_max, dw, pre) self.extractor.dct_norm = self.dct_norm self.extractor.mel_scale = self.mel_scale self.extractor.with_energy = self.withEnergy self.extractor.with_delta = self.withDelta self.extractor.with_delta_delta = self.withDeltaDelta return True def normalize_features(self, params): ######################### ## Initialisation part ## ######################### normalized_vector = [ [ 0 for i in range(params.shape[1]) ] for j in range(params.shape[0]) ] for index in range(params.shape[1]): vector = numpy.array([row[index] for row in params]) n_samples = len(vector) norm_vector = normalize_std_array(vector) for i in range(n_samples): normalized_vector[i][index]=numpy.asscalar(norm_vector[i]) data = numpy.array(normalized_vector) return data def process(self, inputs, outputs): float_wav = inputs["speech"].data.value labels = inputs["vad"].data.value cepstral_features = self.extractor(float_wav) features_mask = self.features_mask filtered_features = numpy.ndarray(shape=((labels == 1).sum(),len(features_mask)), dtype=numpy.float64) i=0 cur_i=0 for row in cepstral_features: if i < len(labels): if labels[i]==1: for k in range(len(features_mask)): filtered_features[cur_i,k] = row[features_mask[k]] cur_i = cur_i + 1 i = i+1 else: if labels[-1]==1: if cur_i == cepstral_features.shape[0]: for k in range(len(features_mask)): filtered_features[cur_i,k] = row[features_mask[k]] cur_i = cur_i + 1 i = i+1 if self.normalizeFeatures: normalized_features = self.normalize_features(filtered_features) else: normalized_features = filtered_features if normalized_features.shape[0] == 0: print("Warning: no speech found in: %s" % input_file) # But do not keep it empty!!! This avoids errors in next steps normalized_features=numpy.array([numpy.zeros(len(features_mask))]) outputs["features"].write({ 'value': numpy.vstack(normalized_features) }) return TrueThe code for this algorithm in Python
The ruler at 80 columns indicate suggested POSIX line breaks (for readability).
The editor will automatically enlarge to accomodate the entirety of your input
Use keyboard shortcuts for search/replace and faster editing. For example, use Ctrl-F (PC) or Cmd-F (Mac) to search through this box
This algorithm implements the MFCC and LFCC feature extraction. It relies on the Bob library.
The following parameters are set inside the algorithm and can be modified by the user: