System and method for neural network based feature extraction for acoustic model development
A system and method are presented for neural network based feature extraction for acoustic model development. A neural network may be used to extract acoustic features from raw MFCCs or the spectrum, which are then used for training acoustic models for speech recognition systems. Feature extraction may be performed by optimizing a cost function used in linear discriminant analysis. General non-linear functions generated by the neural network are used for feature extraction. The transformation may be performed using a cost function from linear discriminant analysis methods which perform linear operations on the MFCCs and generate lower dimensional features for speech recognition. The extracted acoustic features may then be used for training acoustic models for speech recognition systems.
1. A method for training acoustic models in speech recognition systems, wherein the speech recognition system comprises a neural network, the method comprising the steps of:
a. extracting acoustic features from a speech signal using the neural network; and
b. processing the acoustic features into an acoustic model by the speech recognition system,
wherein the neural network comprises at least one of: activation functions with parameters, prealigned feature data, and training,
wherein the training is performed using a stochastic gradient descent method on a cost function, and
wherein the cost function is a linear discriminant analysis cost function.
2. The method of claim 1 , wherein the acoustic features are extracted from Mel Frequency Cepstral Coefficients.
3. The method of claim 1 , wherein the features are extracted from a speech signal spectrum.
4. The method of claim 1 , wherein the processing comprises using at least one of a maximum likelihood method and an expectation maximization method.
5. A method for training acoustic models in speech recognition systems, wherein the speech recognition system comprises a neural network, the method comprising the steps of:
a. extracting acoustic features from a speech signal using the neural network; and
b. processing the acoustic features into an acoustic model by the speech recognition system,
wherein the extracting of step (a) further comprises the step of optimizing a cost function,
wherein the cost function is capable of transforming general non-linear functions generated by the neural network.
6. The method of claim 5 , wherein the transforming comprises:
a. performing non-linear operations on the features; and
b. generating lower dimensional features for speech recognition.
7. The method of claim 6 , wherein the transforming further comprises the generation of Linear Discriminant Analysis transforms, wherein the transforms are generated by optimizing the cost function with linear activation functions.
8. The method of claim 5 , wherein the neural network carries activation functions with variable parameters.
9. The method of claim 8 , wherein the neural network is recurrent.
10. The method of claim 8 , wherein the parameters are determined during training of the neural network.
11. A method for training acoustic models in speech recognition systems, wherein the speech recognition system comprises a neural network, the method comprising the steps of:
a. extracting trainable features from an incoming audio signal using the neural network; and
b. processing the trainable features into an acoustic model by the speech recognition system,
wherein the neural network comprises at least one of: activation functions
with parameters, prealigned feature data, and training,
wherein the training is performed using a stochastic gradient descent
method on a cost function, and
wherein the cost function is a linear discriminant analysis cost function.
12. The method of claim 11 , wherein the trainable features are extracted from Mel Frequency Cepstral Coefficients, wherein the Mel Frequency Cepstral Coefficients are extracted from the audio signal.
13. The method of claim 11 , wherein the processing comprises using at least one of a maximum likelihood method and an expectation maximization method.
14. A method for training acoustic models in speech recognition systems, wherein the speech recognition system comprises a neural network, the method comprising the steps of:
a. extracting trainable features from an incoming audio signal using the neural network; and
b. processing the trainable features into an acoustic model by the speech recognition system,
wherein the extracting of step (a) further comprises the step of optimizing a cost function, wherein the cost function is capable of transforming general non-linear functions generated by the neural network.
15. The method of claim 14 , wherein the neural network carries activation functions with variable parameters.
16. The method of claim 15 , wherein the neural network is recurrent.
17. The method of claim 15 , wherein the parameters are determined during training of the neural network.
18. The method of claim 14 , wherein the transforming comprises:
a. performing non-linear operations on the features; and
b. generating lower dimensional features for speech recognition.
19. The method of claim 18 , wherein the transforming further comprises the generation of Linear Discriminant Analysis transforms, wherein the transforms are generated by optimizing the cost function with linear activation functions.