Speech Recognition Using Dual Acoustic Models

Solution Overview

Problem

Conventional speech command recognition systems have a high false acceptance rate, especially in noisy environments, leading to user frustration and reduced utilization, as they struggle to distinguish between foreground and background acoustic data.

Innovation Solution

The method employs a dual acoustic model approach, using a background acoustic model with low precision and a foreground acoustic model with high precision to differentiate between sounds originating from the user and background noise, assigning weights to sound samples based on their likelihood of originating from the foreground, thereby reducing false acceptance rates.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional speech command recognition methods are used, then the system can process speech input, but the false acceptance rate becomes unacceptably high in noisy environments

Engineering Contradiction:

Improvefalse acceptance rateVSAvoidbackground noise interference

Core Design Contradiction:

ReliabilityVSObject-affected harmful factors

Solution Approach 1:

The acoustic model is segmented into two distinct components: a foreground acoustic model trained on clean speech data and a background acoustic model trained on noisy environment data. This segmentation allows the system to separately model and distinguish between foreground speech signals and background noise, thereby reducing false acceptance rates in noisy environments while maintaining reliable speech command recognition

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The system dynamically adjusts the weight parameter (first weight) assigned to the foreground acoustic model based on the computed probability that speech originated from the foreground. This parameter change enables adaptive optimization of recognition accuracy across varying noise conditions, improving reliability without being overly sensitive to background noise

Inventive Principle:

Principle #35Parameter changes

2Measurement precision

If a single acoustic model is used for speech recognition, then the device complexity is low, but the measurement precision of speech commands deteriorates in noisy environments

Engineering Contradiction:

Improvespeech command recognition accuracyVSAvoidacoustic model structure

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The acoustic model is segmented into two distinct components: a foreground acoustic model trained on clean speech data and a background acoustic model trained on noisy environment data. This segmentation allows the system to separately model and distinguish between foreground speech signals and background noise, thereby reducing false acceptance rates in noisy environments while maintaining reliable speech command recognition

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The dual acoustic model system serves multiple functions: the foreground model handles clean speech recognition, the background model handles noisy environment characterization, and their combination through probability-weighted decoding provides adaptive speech recognition across varying conditions. This multi-functionality achieves high measurement precision without proportionally increasing device complexity

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Measurement precision

If the system uses a foreground acoustic model with high precision, then speech command recognition accuracy improves, but the computational resources and energy consumption increase

Engineering Contradiction:

Improvespeech command recognition accuracyVSAvoidenergy consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The system applies partial action by selectively engaging the computationally intensive foreground acoustic model only when the probability of foreground speech exceeds a threshold (i.e., when first weight indicates likely foreground origin). For low-probability cases, the system relies on the lighter background model, thereby reducing overall energy consumption while maintaining high recognition accuracy for genuine speech commands

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The system dynamically adjusts the weight parameter (first weight) assigned to the foreground acoustic model based on the computed probability that speech originated from the foreground. This parameter change enables adaptive optimization of recognition accuracy across varying noise conditions, improving reliability without being overly sensitive to background noise

Inventive Principle:

Principle #35Parameter changes

Data Source

PatentUS9805715B2Method and system for recognizing speech commands using background and foreground acoustic models

Publication Date: 2017.10.31 TENCENT TECHNOLOGY (SHENZHEN) CO LTD

AI summary

A method of recognizing speech commands includes generating a background acoustic model for a sound using a first sound sample, the background acoustic model characterized by a first precision metric. A foreground acoustic model is generated for the sound using a second sound sample, the foreground acoustic model characterized by a second precision metric. A third sound sample is received and decoded by assigning a weight to the third sound sample corresponding to a probability that the sound sample originated in a foreground using the foreground acoustic model and the background acoustic model. The method further includes determining if the weight meets predefined criteria for assigning the third sound sample to the foreground and, when the weight meets the predefined criteria, interpreting the third sound sample as a portion of a speech command. Otherwise, recognition of the third sound sample as a portion of a speech command is forgone.