Compressed Gaussian Probability Data Bit Reduction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition algorithms require significant memory resources due to the large number of floating-point operations and parameters involved in computing Gaussian probabilities, leading to a considerable demand on available memory without sacrificing recognition accuracy.
Innovation Solution
The use of compressed probability distribution functions with α-bit floating point numbers for mean and variance values, converted to β-bit integers, reduces memory usage while maintaining recognition accuracy by rewriting Gaussian functions with compressed mean and variance values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If 32-bit floating point numbers are used for Gaussian mean and variance values, then computation precision is maintained, but memory usage becomes excessively large
Solution Approach 1:
The patent changes the data representation parameters from 32-bit floating point numbers to 8-bit integers for storing Gaussian mean and variance values. This parameter transformation reduces memory usage significantly while maintaining sufficient precision for speech recognition computations, directly resolving the contradiction between precision and memory consumption.
Solution Approach 2:
The patent creates a compressed copy of the Gaussian parameters using 8-bit integers that approximate the original 32-bit floating point values. This compressed copy is then used in the speech recognition algorithm, achieving memory reduction while preserving the essential computational functionality and recognition accuracy.
2Measurement precision
If the number of Gaussians is increased to improve recognition accuracy, then computational accuracy improves, but memory demand increases considerably
Solution Approach 1:
The patent applies parameter changes by representing each Gaussian with compressed 8-bit integer values instead of full 32-bit floats. This allows the system to store a larger number of Gaussians in memory while maintaining manageable memory consumption, thereby enabling improved recognition accuracy through increased Gaussian count without proportionally increasing memory demand.
Data Source
AI summary
Use of runtime memory may be reduced in a data processing algorithm that uses one or more probability distribution functions. Each probability distribution function may be characterized by one or more uncompressed mean values and one or more variance values. The uncompressed mean and variance values may be represented by α-bit floating point numbers, where α is an integer greater than 1. The probability distribution functions are converted to compressed probability functions having compressed mean and/or variance values represented as β-bit integers, where β is less than α, whereby the compressed mean and/or variance values occupy less memory space than the uncompressed mean and/or variance values. Portions of the data processing algorithm can be performed with the compressed mean and variance values.