Localized Speech Recognition Offload for Speaker Adaptation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing automatic speech recognition (ASR) systems face challenges in providing efficient speech recognition on devices with limited computational resources and require either device-specific ASR modules or remote network access, which can be impractical for simple devices and may not account for speaker-specific vocal characteristics.
Innovation Solution
A local ASR system is deployed that offloads speech recognition tasks from devices, supporting both text and command-based responses, and performs speaker adaptation to improve recognition accuracy across various devices without the need for extensive computational resources on user devices.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If device-specific ASR modules are deployed on each user device, then speech recognition accuracy for that device is improved, but device complexity and computational resource requirements increase
Solution Approach 1:
The patent divides the ASR functionality into two segments: a centralized ASR server that handles speech recognition processing and device-specific adaptation, and simple user devices that only need to capture audio and transmit it to the server. This segmentation allows high accuracy through device-specific adaptation without requiring complex ASR modules on each device.
Solution Approach 2:
The ASR server acts as an intermediary between the user device and the speech recognition processing. The server receives audio from simple user devices, performs complex speech recognition and speaker adaptation, then provides appropriate responses. This intermediary approach eliminates the need for complex ASR functionality on the user devices themselves.
2Adaptability or versatility
If ASR processing is performed on devices with limited computational resources, then device versatility is improved, but speech recognition accuracy deteriorates
Solution Approach 1:
The patent extracts the computationally intensive ASR processing functionality from the user devices and places it in a centralized server. The user devices only retain the simple audio capture and transmission functions, which can be implemented on devices with limited computational resources. The server handles all complex speech recognition and speaker adaptation tasks.
3Measurement precision
If speaker adaptation is performed on each device, then speech recognition accuracy for that speaker is improved, but communication overhead with remote systems increases
Solution Approach 1:
The patent merges speaker adaptation functionality with the centralized ASR server processing. The server maintains speaker-specific acoustic models and adaptation parameters, combining multiple functions (speech recognition, speaker identification, and adaptation) into a single centralized processing location. This eliminates redundant communication overhead that would occur if each device maintained its own adaptation models.
Data Source
AI summary
A local computing device may receive an utterance from a user device. In response to receiving the utterance, the local computing device may obtain a text string transcription of the utterance, and determine a response mode for the utterance. If the response mode is a text-based mode, the local computing device may provide the text string transcription to a target device. If the response mode is a non-text-based mode, the local computing device may convert the text string transcription into one or more commands from a command set supported by the target device, and provide the one or more commands to the target device.