Hot Word Detection for Dynamic Speech Recognition Channel Allocation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Speech recognition processors are inefficient in detecting relevant words in a stream of irrelevant words during conversations, leading to high resource usage and costly implementation in on-demand voice-enabled applications, as they require dedicated channels and are prone to false positives.
Innovation Solution
A speech analytics processor monitors conversations for a 'hot word' to invoke a speech recognition processor only when needed, allowing the recognition processor to be decoupled and reused among multiple voice sessions, reducing resource allocation and costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If a speech recognition processor is dedicated to a voice session for the entire session, then the speech recognition processor can provide continuous speech processing services, but the resource consumption increases and cannot support hundreds or thousands of voice sessions concurrently
Solution Approach 1:
The system dynamically allocates speech recognition processor channels based on real-time detection of hot words in conversations. The speech analytics processor continuously monitors conversations and only activates the speech recognition processor when a hot word is detected, allowing the system to adapt resource allocation from static dedicated channels to dynamic on-demand channels, thereby supporting more concurrent voice sessions with limited processor resources
Solution Approach 2:
The speech analytics processor performs preliminary monitoring of conversations to detect hot words before activating the speech recognition processor. This preliminary detection action enables the system to prepare and activate speech recognition resources only when needed, rather than maintaining them continuously, thus reducing overall resource consumption while ensuring availability when required
2Ease of operation
If a speech recognition processor attempts to match each word in a conversation to a predefined grammar, then the processor can provide speech recognition services, but it generates false positive results and error indicators for irrelevant words
Solution Approach 1:
The speech analytics processor acts as an intermediary between the conversation stream and the speech recognition processor. It pre-processes the conversation by detecting hot words and filtering out irrelevant content, then only passes relevant segments to the speech recognition processor. This intermediary function prevents the speech recognition processor from attempting to match every word against the grammar, thereby eliminating false positives while maintaining accurate recognition for relevant commands
Solution Approach 2:
The system extracts and isolates only the relevant portions of conversation (those containing hot words) for speech recognition processing. By separating the conversation stream into relevant and irrelevant segments and only submitting relevant segments to the speech recognition processor, the system eliminates unnecessary processing and false matches while preserving the ability to recognize commands accurately
3Adaptability or versatility
If speech recognition processors are used in on-demand voice-enabled applications, then voice-enabled functionality can be provided, but the high memory and processing requirements make simultaneous support for hundreds or thousands of voice sessions impractical or cost-prohibitive
Solution Approach 1:
Instead of continuously maintaining speech recognition processor channels for all voice sessions, the system uses periodic action by only activating the speech recognition processor when hot words are detected in conversations. The speech analytics processor periodically monitors conversations and triggers speech recognition processing only at those specific moments, transforming the resource consumption pattern from continuous to periodic, thereby reducing overall energy and resource usage while maintaining on-demand functionality
Solution Approach 2:
The system replaces expensive, long-lived dedicated speech recognition processor channels with a more economical model using a small pool of shared processors that are activated temporarily only when needed. The speech recognition processor channels are allocated and deallocated dynamically based on hot word detection, effectively treating them as short-lived resources rather than permanent dedicated channels,ไป่ significantly reducing the total number of processors needed
Data Source
AI summary
A method and apparatus for detecting use of an utterance. A voice session including voice signals generated during a conversation between a first participant and a second participant is monitored by a speech analytics processor. The speech analytics processor detects the use of an utterance. A speech recognition processor channel selected from a pool of speech recognition processor channels and is coupled to the voice session. The speech recognition processor provided speech recognition services to a voice-enabled application. The speech recognition processor channel is then decoupled from the voice session. The speech analytics processor continues to monitor the conversation for subsequent use of the utterance.