Sound Classification for Speech Recognition in Noisy Environments
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Voice-driven systems face challenges in distinguishing intended speech from extraneous sounds, particularly in noisy environments like warehouses, where public address announcements can be mistaken for user input due to their similar aural qualities.
Innovation Solution
A sound classification system that differentiates between speech, non-transient background noise, and transient noise events, using a headset with multiple microphones to classify audio frames and provide confidence values to a processor-based speech recognition device, which adjusts thresholds to prevent false positives.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the speech recognition system uses a single threshold for speech detection, then the system is simple to operate, but it cannot distinguish between speech and transient noise events like public address announcements
Solution Approach 1:
The patent segments the audio signal into discrete frames and classifies each frame independently as speech, non-transient background noise, or transient noise event. This frame-by-frame segmentation allows the system to apply different handling strategies to different types of sounds, improving detection precision without requiring a monolithic complex system
Solution Approach 2:
The patent changes the parameter of noise detection by introducing a metric that compares current frame noise levels to a running average of past noise levels. When the current noise exceeds the average by a threshold, the frame is classified as a transient noise event. This parameter-based approach enables automatic distinction between speech and transient noises like public address announcements
2Reliability
If the system classifies all noise as background noise, then the classification process is simple, but transient noise events are mistakenly recognized as speech
Solution Approach 1:
The patent performs preliminary classification of audio frames into three categories (speech, non-transient background noise, transient noise events) before the speech recognition process. By pre-identifying transient noise events and excluding them from speech recognition processing, the system prevents false positives without interfering with the main speech recognition functionality
Solution Approach 2:
The patent introduces an intermediary classification layer between the microphone input and the speech recognition engine. This intermediary component analyzes audio frames, determines their type, and selectively passes only speech frames to the speech recognition system. This intermediary prevents transient noise events from being misrecognized as speech while maintaining system reliability
3Productivity
If the system processes all audio frames as potential speech, then no speech is missed, but processing time and computational resources increase
Solution Approach 1:
The patent extracts and removes transient noise event frames from the audio stream before passing remaining frames to speech recognition. By taking out frames classified as transient noise events (such as public address announcements), the system reduces the number of frames requiring computationally intensive speech recognition processing, thereby improving processing speed without losing actual speech information
Solution Approach 2:
The patent applies partial action by processing only those audio frames that are classified as speech or potential speech through the full speech recognition pipeline. Frames classified as non-transient background noise or transient noise events receive lighter processing or are excluded entirely. This selective processing approach maintains speech detection completeness while significantly improving processing efficiency
Data Source
Figure 1
Figure 2
Figure 3
AI summary
Audio frames are classified as either speech, non-transient background noise, or transient noise events. Probabilities of speech or transient noise event, or other metrics may be calculated to indicate confidence in classification. Frames classified as speech or noise events are not used in updating models (e.g., spectral subtraction noise estimates, silence model, background energy estimates, signal-to-noise ratio) of non-transient background noise. Frame classification affects acceptance/rejection of recognition hypothesis. Classifications and other audio related information may be determined by circuitry in a headset, and sent (e.g., wirelessly) to a separate processor-based recognition device.