ASR Segmentation Reduces Computational Load on Edge Devices
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing speech recognition systems that use neural networks, such as RNN-T models, require substantial computation resources, which can be a challenge for implementation on user devices with limited resources, especially when privacy concerns necessitate local processing.
Innovation Solution
The proposed solution involves performing automatic speech recognition (ASR) processing in an end-to-end manner on the user device using an encoder, segmenter, segment aggregator, and decoder, which allows for segmenting speech into defined chunks, encoding each variable-length segment into a fixed-length representation, and decoding these segments independently to generate word sequences.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural network-based speech recognition (e.g., RNN-T models) is implemented, then speech recognition accuracy is improved, but computational resource requirements increase
Solution Approach 1:
The speech signal is divided into fixed-duration audio frames that are processed sequentially through the neural network. This segmentation allows the system to process speech in manageable chunks, reducing the computational burden compared to processing the entire speech signal at once, while still maintaining accurate recognition through the sequential analysis of frames
2Reliability
If speech processing is performed locally on user devices, then user privacy is protected, but device computational resources are strained
Solution Approach 1:
By segmenting speech into fixed-duration frames and processing them through an optimized neural network architecture, the system reduces the computational complexity required for local processing. This enables privacy-protecting local inference on resource-constrained devices without requiring excessive computational resources
Solution Approach 2:
The system transforms the continuous speech signal into discrete fixed-duration audio frames with specific parameter constraints. This parameterization approach simplifies the processing requirements while maintaining the essential information needed for accurate speech recognition, making local processing feasible on mobile devices
Data Source
AI summary
Techniques for performing automatic speech recognition (ASR) processing are described. The ASR processing may involve use of a segmenter and a decoder. The segmenter may be configured to identify audio segments, from audio data containing speech, based on word boundaries. The decoder may be configured to generate multiple word hypotheses for individual audio segments. A fixed-size representation of an audio segment may be generated. The described ASR processing techniques may be computationally less expensive than at least some other systems. Also, the described ASR processing techniques may maintain a larger number of word predictions per audio segment as compared to at least some other systems.