Speaker-Verified Session Management for Speech Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional speech recognition systems require users to repeatedly activate sessions for each utterance, leading to inconvenient operation and poor performance, as they cannot distinguish between intended speech and noise, resulting in unnecessary power consumption and incorrect speech recognition.
Innovation Solution
Implementing a speech recognition system that maintains a session based on speaker recognition, allowing sequential speech recognition without additional activation, and outputs results only for registered speakers, thereby providing personalized service and improving user convenience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If the system requires users to repeatedly activate sessions for each utterance, then the speech recognition can be performed, but the operation becomes inconvenient and user experience deteriorates
Solution Approach 1:
The system performs preliminary speaker recognition to determine whether to maintain the session before speech recognition is needed. By pre-establishing speaker identity and session status, the system eliminates the need for users to manually activate sessions for each utterance, thereby improving operational convenience and reducing time loss.
2Reliability
If the system performs speech recognition on all audio inputs, then no speech is missed, but power consumption increases and unnecessary processing occurs
Solution Approach 1:
The system extracts and processes only the relevant portion of audio inputs by first performing speaker recognition. When a non-registered speaker is detected, the system extracts and discards the audio signal without performing full speech recognition, thereby reducing unnecessary processing and power consumption while maintaining reliable speech recognition for registered speakers.
3Productivity
If the system cannot distinguish between intended speech and noise, then all audio is processed, but recognition accuracy decreases and incorrect results occur
Solution Approach 1:
The system introduces speaker recognition as an intermediary step between audio input and speech recognition. This intermediary process identifies whether the audio signal comes from a registered speaker, allowing the system to selectively process only relevant speech inputs and filter out noise or unintended inputs, thereby improving recognition accuracy while maintaining efficient processing throughput.
4Adaptability or versatility
If the system maintains sessions for all speakers, then sequential speech recognition is enabled, but resource usage increases and personalization is lost
Solution Approach 1:
The system applies different session management policies to different speakers based on their registration status. Registered speakers receive continuous session maintenance enabling sequential speech recognition, while non-registered speakers do not trigger session maintenance. This localized quality approach optimizes resource usage by personalizing session handling according to speaker identity.
Data Source
AI summary
A speech recognition method and an apparatus which recognize speech, based on speaker recognition, and output a result of the speech recognition are provided. The speech recognition method includes activating a session for receiving an input of an audio signal, performing speech recognition on a speech signal detected from the input audio signal while the session is maintained, determining whether a speaker of the speech signal is a registered speaker based on speaker information generated from the speech signal, determining whether to maintain the session based on a result of the determination, and outputting a result of performing the speech recognition.