Variable-Component DNN for Robust Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face challenges in maintaining consistent quality across varying environmental conditions, such as different noise levels and locations, due to their inability to effectively model and adapt to diverse acoustic environments.
Innovation Solution
The implementation of a variable-component deep neural network (VCDNN) that models almost any component as a set of polynomial functions of environment variables, including signal-to-noise ratio, speaking rate, and user characteristics, allowing for dynamic adaptation during speech recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a standard deep neural network is used for speech recognition, then the system structure is simple and easy to implement, but the recognition quality deteriorates in varying environmental conditions
Solution Approach 1:
The patent applies dynamics by making the neural network components (weighting matrices and bias components) variable rather than fixed. The components dynamically adapt based on environment variables such as signal-to-noise ratio, allowing the network to adjust its behavior according to acoustic conditions while maintaining a relatively simple overall structure
Solution Approach 2:
The patent changes parameters by modeling the weighting matrix and bias components as polynomial functions of environment variables. This allows the network to incorporate environmental information through parameter variation rather than structural complexity, improving recognition quality across different acoustic environments
2Adaptability or versatility
If environment variables are incorporated into the DNN components, then speech recognition quality improves across multiple environments, but the computational complexity increases
Solution Approach 1:
The patent models DNN components as polynomial functions of environment variables, allowing the network to adapt to different acoustic conditions through parameter variation. This approach enables environmental adaptability while keeping the computational overhead manageable through efficient polynomial evaluation
Solution Approach 2:
The patent pre-computes and stores polynomial coefficients for the environment variable relationships during training. During inference, these pre-computed coefficients are used to quickly evaluate the variable components, reducing real-time computational complexity while maintaining adaptability
Data Source
AI summary
Systems and methods for speech recognition incorporating environmental variables are provided. The systems and methods capture speech to be recognized. The speech is then recognized utilizing a variable component deep neural network (DNN). The variable component DNN processes the captured speech by incorporating an environment variable. The environment variable may be any variable that is dependent on environmental conditions or the relation of the user, the client device, and the environment. For example, the environment variable may be based on noise of the environment and represented as a signal-to-noise ratio. The variable component DNN may incorporate the environment variable in different ways. For instance, the environment variable may be incorporated into weighting matrices and biases of the DNN, the outputs of the hidden layers of the DNN, or the activation functions of the nodes of the DNN.