Skip to content

Module: am.SpeakerVerification

Overview

Classes

class MFCC

class SpeakerVerificationModel

am.SpeakerVerification.MFCC

Defined in animius/SpeakerVerification/MFCC.py

Methods

get_MFCC

get_MFCC(path, window=10, step=3, num_cepstral=39, flatten=False)

Get MFCC data.

Args:

  • path (str) -- path to wav files.

  • window (int) -- amount of window. (Optional)

  • step (int) -- amount of step. (Optional)

  • num_cepstral (int) -- amount of cepstral. (Optional)

  • flatten (boolean) -- whether or not to output flatten data. (Optional)

Returns:

A numpy array of MFCC data.

am.SpeakerVerification.SpeakerVerificationModel

Defined in animius/SpeakerVerification/SpeakerVerificationModel.py

__init__

Args: None

Methods

DEFAULT_HYPERPARAMETERS (Static Method)

am.SpeakerVerification.SpeakerVerificationModel.DEFAULT_HYPERPARAMETERS()

Get default hyperparameters of SpeakerVerification model.

Args: None

Returns:

1
2
3
4
5
{
    'learning_rate': 0.005,
    'batch_size': 2048,
    'optimizer': 'adam'
}

DEFAULT_MODEL_STRUCTURE (Static Method)

am.SpeakerVerification.SpeakerVerificationModel.DEFAULT_MODEL_STRUCTURE()

Get default model structure of SpeakerVerification model.

Args: None

Returns:

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
{
    'filter_size_1': 3,
    'num_filter_1': 10,
    'pool_size_1': 2,
    'pool_type': 'max',
    'filter_size_2': 5,
    'num_filter_2': 15,
    'fully_connected_1': 128,
    'input_window': 10,
    'input_cepstral': 39
}

build_graph

build_graph(model_config, data, graph=None)

Build the graph for SpeakerVerification model.

Args:

  • model_config (am.ModelConfig) -- reference to an am.ModelConfig object.

  • data (am.SpeakerVerificationData) -- reference to an am.SpeakerVerificationData object.

  • graph (tf.Graph) -- reference to a tf.Graph object.

Returns: None

train

train(epochs=800, CancellationToken=None)

Train model with specific epochs.

Args:

  • epochs (int) -- amount of epochs which the model will train with. (Optional)

  • data (am.SpeakerVerificationData) -- reference to an am.SpeakerVerificationData object.

  • CancellationToken (NoneType) -- CancellationToken. (Optional)

Returns: None

load (Class method)

load(directory, name='model', data=None)

Load SpeakerVerification model from local files.

Args:

  • directory (str) -- directory of files.

  • name (str) -- name of SpeakerVerification model.

  • data (am.SpeakerVerificationData) -- reference to an am.SpeakerVerificationData object.

Returns:

The reference to an am.SpeakerVerification.SpeakerVerificationModel object.

predict

predict(input_data, save_path=None, raw=False)

Predict input data and save predicted values to local files.

Args:

  • input_data (str) -- input data.

  • save_path (str) -- path to save results. (Optional)

  • raw (boolean) -- whether or not to return raw data. (Optional)

Returns:

Predicted data.

predict_folder

predict_folder(input_data, folder_directory, save_path=None, raw=False)

Read data in the folder and predict results.

Args:

  • input_data (str) -- input data.

  • folder_directory (str) -- path to save results. (Optional)

  • save_path (str) -- path to save results. (Optional)

  • raw (boolean) -- whether or not to return raw data. (Optional)

Returns:

Predicted data.