Adapting ASR Parameters via Hotword Attributes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automated speech recognition systems face challenges in accurately recognizing short, noisy, and open-ended utterances in multi-user environments, where context and user-specific preferences are crucial for improving recognition accuracy.
Innovation Solution
The method involves extracting hotword attributes from detected hotwords in streaming audio, such as pause duration, rate of speech, pitch, and confidence scores, to adjust speech recognition parameters, including beam search width and hypothesis output, and applying frequency-based filtering to optimize speech recognition results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the ASR model processes all audio segments with full recognition resources, then speech recognition accuracy is improved, but energy consumption and processing time increase
Solution Approach 1:
The patent dynamically adjusts ASR processing parameters (such as beam search width, hypothesis output count, and model complexity) based on hotword attributes like confidence scores, pause durations, and speech rates. When confidence is high or pause duration is short, the system reduces processing resources; when confidence is low or pause duration is long, it increases resources, thereby optimizing the balance between accuracy and energy consumption
Solution Approach 2:
The system transitions from a static ASR processing approach to a dynamic one where processing intensity adapts in real-time based on extracted hotword attributes. The ASR model selectively applies different levels of processing complexity depending on the situation, making the system more efficient while maintaining accuracy when needed
2Productivity
If the ASR model uses default processing parameters for all segments, then processing speed is maintained, but speech recognition accuracy decreases in challenging conditions
Solution Approach 1:
The patent modifies ASR processing parameters based on hotword attributes such as confidence scores, pause durations, and speech rates. When attributes indicate challenging conditions (low confidence, long pauses, unusual speech rates), the system increases processing resources and adjusts parameters to improve accuracy; otherwise, it maintains default parameters for efficient processing
Solution Approach 2:
The system performs preliminary analysis of hotword attributes before initiating full ASR processing. By extracting and evaluating attributes like confidence scores and pause durations in advance, the system can pre-determine the appropriate processing level, avoiding unnecessary complex processing while ensuring accuracy when needed
3Loss of information
If the system captures complete user queries with extended endpointing timeouts, then query completeness is improved, but processing time and energy consumption increase
Solution Approach 1:
The patent adjusts endpointing timeout parameters based on extracted hotword attributes such as pause duration and speech rate. When pause duration is short or speech rate is normal, the system uses shorter timeouts to reduce processing time; when pause duration is long or speech rate is slow, it extends timeouts to ensure complete query capture, thereby dynamically optimizing the balance between completeness and efficiency
Data Source
AI summary
A method for optimizing speech recognition includes receiving a first acoustic segment characterizing a hotword detected by a hotword detector in streaming audio captured by a user device, extracting one or more hotword attributes from the first acoustic segment, and adjusting, based on the one or more hotword attributes extracted from the first acoustic segment, one or more speech recognition parameters of an automated speech recognition (ASR) model. After adjusting the speech recognition parameters of the ASR model, the method also includes processing, using the ASR model, a second acoustic segment to generate a speech recognition result. The second acoustic segment characterizes a spoken query/command that follows the first acoustic segment in the streaming audio captured by the user device.