Keyword Index Refinement via Speech Rate Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems face challenges in efficiently processing audio samples due to the need for always-on power usage, which reduces battery life in mobile devices and increases processing demands.
Innovation Solution
A method and apparatus for processing audio samples that involves detecting spoken keywords using a first keyword detection model, determining estimated keyword indices, classifying speech rate using a machine learning network, obtaining an average spoken length value, and generating refined keyword indices to improve detection accuracy and reduce latency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If speech recognition is implemented as an always-on function to maximize utility, then detection accuracy and responsiveness are improved, but power consumption increases and battery life is reduced
Solution Approach 1:
The speech recognition system is divided into multiple stages: a first keyword detection model performs initial detection, and a second keyword detection model performs refined detection only when needed. This segmentation allows the system to maintain high detection accuracy while reducing overall power consumption by activating the more resource-intensive second model only for promising candidates rather than continuously processing all audio.
Solution Approach 2:
The first keyword detection model performs preliminary filtering of audio samples before they reach the second keyword detection model. By pre-processing and identifying potential keywords in advance, the system reduces the computational burden on the second model and minimizes the time both models need to remain active, thereby reducing power consumption while maintaining detection accuracy.
2Measurement precision
If keyword detection processing is extended to improve accuracy, then detection performance is improved, but processing latency increases
Solution Approach 1:
The system performs preliminary keyword detection using the first model to identify candidate keywords and their approximate time ranges before applying the more accurate second model. This preliminary action provides the second model with focused input, reducing its processing time while maintaining high detection accuracy through the two-stage approach.
Solution Approach 2:
The second keyword detection model is applied selectively only to audio segments containing detected keywords rather than processing the entire audio stream uniformly. This localized application of high-quality processing improves keyword detection accuracy for critical segments while minimizing overall processing latency by avoiding unnecessary computation in non-relevant audio portions.
Data Source
AI summary
Systems and techniques are provided for processing one or more audio samples. For example, a process can include detecting, using a first keyword detection model, a spoken keyword within an audio sample of the one or more audio samples. Estimated keyword indices corresponding to detection of the spoken keyword within the audio sample can be determined, comprising an estimated keyword start index and an estimated keyword end index. A speech rate classification machine learning network can be used to determine speech rate information corresponding to the audio sample. An average spoken length value corresponding to the spoken keyword and the speech rate information can be obtained. Refined keyword indices can be generated based on the estimated keyword indices and the average spoken length value, wherein the refined keyword indices include a refined keyword start index shifted to a time earlier than the estimated keyword start index