BEAT - tutorial/gmm_projection/3

This algorithm is a legacy one. The API has changed since its implementation. New versions and forks will need to be updated.

This algorithm is splittable

Endpoint Groups 2

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.

Group: main

Endpoint Name	Data Format	Nature
features	system/array_2d_floats/1	Input
id	system/uint64/1	Input
model	tutorial/gmm/1	Output

Unnamed group

Endpoint Name	Data Format	Nature
ubm	tutorial/gmm/1	Input

import bob
import numpy
from bob.machine import GMMMachine

def gmm_from_data(data):
    """Unmangles a bob.machine.GMMMachine from a BEAT Data object"""

dim_c, dim_d = data.means.shape
    gmm = GMMMachine(dim_c, dim_d)
    gmm.weights = data.weights
    gmm.means = data.means
    gmm.variances = data.variances
    gmm.variance_thresholds = data.variance_thresholds
    return gmm

class Algorithm:

def __init__(self):
        self.ubm      = None
        self.features = []

def process(self, inputs, outputs):

# retrieve the UBM once
        if self.ubm is None:
            inputs['ubm'].next()
            self.ubm = gmm_from_data(inputs['ubm'].data)

# collect all the features for the current template
        self.features.append(inputs["features"].data.value)

# adapts the UBM GMM for the template (when all the features have been collected)
        if inputs["id"].isDataUnitDone():
            model = bob.machine.GMMMachine(self.ubm)

trainer = bob.trainer.MAP_GMMTrainer()
            trainer.max_iterations = 1
            trainer.set_prior_gmm(self.ubm)
            trainer.train(model, numpy.vstack(self.features))

# outputs data
            outputs["model"].write({
                'weights':              model.weights,
                'means':                model.means,
                'variances':            model.variances,
                'variance_thresholds':  model.variance_thresholds,
            })

self.features = []

return True

The 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

For a given set of feature vectors and a Gaussian Mixture Models (GMM), this algorithm implements the Maximum-a-posteriori (MAP) estimation (adapting only the means).

Details of MAP estimation can be found in [Reynolds2000]. A very good description on how the MAP estimation works can be found in the Mathematical Monks's YouTube channel.z

This algorithm relies on the Bob library.

The inputs are:

features: A set of floating point vectors as a two-dimensional array (64 bits) of a client. The number of rows correspond to the number of samples, and the number of columns to the dimensionality of the samples.
ubm: A GMM corresponding to the Universal Background Model.
id: Client (class/subject) identifier as an unsigned 64 bits integer.

The output, model, is the adapted GMM (MAP adaptation).

[Reynolds2000]

Reynolds, Douglas A., Thomas F. Quatieri, and Robert B. Dunn. "Speaker verification using adapted Gaussian mixture models." Digital signal processing 10.1 (2000): 19-41.

Experiments

Attestation:

Privacy:

Status:

Name	Databases/Protocols	Analyzers
tutorial/tutorial/full_ubmgmm/2/mobioMale_gmm_DCT12x8_100G	mobio/1@male	tutorial/eerhter_postperf_iso/1
tutorial/tutorial/full_ubmgmm/2/mobioMale_ubmgmm_DCT12x8_100G	mobio/1@male	tutorial/eerhter_postperf_iso/1
tutorial/tutorial/full_ubmgmm/2/bancaP_gmm_DCT12x8_100G	banca/1@P	tutorial/eerhter_postperf_iso/1

This table shows the number of times this algorithm has been successfully run using the given environment. Note this does not provide sufficient information to evaluate if the algorithm will run when submitted to different conditions.

algorithms tutorial gmm_projection 3

Endpoint Groups 2

Group: main

Unnamed group

Experiments

algorithms

tutorial

gmm_projection

3