Conversation Context Model for Speech Recognition Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current speech-driven systems require explicit user activation and manual microphone control, leading to cumbersome interactions and misrecognitions due to unintended audio inputs, hindering their adoption and effectiveness in human-machine and human-human conversations.
Innovation Solution
Implementing a conversation context model that continuously monitors speech and uses n-gram distributions and perplexity scores to determine whether audio inputs are relevant to the ongoing conversation, allowing the system to automatically distinguish between intended and unintended inputs without the need for explicit user signaling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the microphone is activated continuously to capture all user inputs, then the system can respond more naturally without manual activation, but the system receives unintended audio inputs from other people in the vicinity causing misrecognitions
Solution Approach 1:
The patent introduces a conversation context model as an intermediary between the continuous audio input and the speech recognition system. This model analyzes the audio stream and determines whether each utterance is part of the ongoing conversation or from another source, filtering out unintended inputs before they reach the speech recognizer and preventing misrecognitions while maintaining continuous listening capability
Solution Approach 2:
The system uses feedback from the conversation context model to dynamically control speech recognition processing. When the model determines an utterance is relevant to the ongoing conversation, the system activates speech recognition; when it determines the utterance is from another source, the system suppresses recognition, thereby maintaining accuracy while operating in continuous listen mode
2Reliability
If the system waits for explicit user activation before processing speech, then it avoids processing unintended audio inputs, but it requires manual control and cannot respond naturally to spontaneous speech
Solution Approach 1:
The conversation context model operates autonomously to determine whether audio inputs are relevant to the ongoing conversation. It self-manages the filtering process by analyzing contextual cues and automatically deciding which utterances should be processed, eliminating the need for manual activation while maintaining reliable filtering of unintended inputs
Solution Approach 2:
The system performs preliminary analysis of audio inputs through the conversation context model before committing to full speech recognition processing. This preliminary action determines relevance based on contextual information, allowing the system to prepare for natural response while avoiding processing of unrelated audio, thus bridging the gap between automatic operation and accurate filtering
Data Source
AI summary
Disclosed herein are systems, methods, and non-transitory computer-readable storage media for processing speech. A system configured to practice the method monitors user utterances to generate a conversation context. Then the system receives a current user utterance independent of non-natural language input intended to trigger speech processing. The system compares the current user utterance to the conversation context to generate a context similarity score, and if the context similarity score is above a threshold, incorporates the current user utterance into the conversation context. If the context similarity score is below the threshold, the system discards the current user utterance. The system can compare the current user utterance to the conversation context based on an n-gram distribution, a perplexity score, and a perplexity threshold. Alternately, the system can use a task model to compare the current user utterance to the conversation context.