Pre-Wakeword Speech Processing With Reverse ASR for Command Detection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Conventional voice virtual assistants struggle to recognize commands spoken before a wakeword due to the limitations of existing buffering and speech detection methods, leading to inaccurate processing and increased power consumption.

Innovation Solution

Implementing reverse automatic speech recognition (R-ASR) to process speech audio in the reverse order, starting from the wakeword, to distinguish between command and non-command speech, reducing the need for continuous voice activity detection and minimizing power consumption.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If continuous voice activity detection is used to detect commands in pre-wakeword speech, then command detection capability is improved, but power consumption increases

Engineering Contradiction:

Improvecommand detection capabilityVSAvoidpower consumption

Core Design Contradiction:

Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies reverse automatic speech recognition (R-ASR) to process pre-wakeword speech from the wakeword backward in time, rather than continuously forward. This inversion allows the system to stop processing as soon as a complete command is identified, eliminating the need for continuous voice activity detection and reducing power consumption while maintaining command detection capability.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The system performs preliminary buffering of speech audio before the wakeword, storing a fixed duration of audio in memory. This preliminary action allows the system to have immediate access to pre-wakeword speech without continuous processing, enabling efficient reverse processing only when needed upon wakeword detection.

Inventive Principle:

Principle #10Preliminary action

2Ease of manufacture

If a fixed duration buffer is used to store pre-wakeword speech, then implementation simplicity is improved, but accuracy of command identification deteriorates due to inclusion of non-command speech

Engineering Contradiction:

Improveimplementation simplicityVSAvoidcommand identification accuracy

Core Design Contradiction:

Ease of manufactureVSMeasurement precision

Solution Approach 1:

By processing the buffered speech in reverse order from the wakeword backward, the system can identify complete commands more accurately. The reverse processing allows the system to stop as soon as a complete command is found, avoiding inclusion of preceding non-command speech that would contaminate the command identification in forward processing.

Inventive Principle:

Principle #13The other way round (Inversion)

Solution Approach 2:

The system extracts only the relevant command portion from the buffered pre-wakeword speech by processing backward from the wakeword and stopping when a complete command is identified. This extraction eliminates unnecessary non-command speech from processing, improving command identification accuracy while maintaining implementation simplicity.

Inventive Principle:

Principle #2Taking out (Extraction)

3Adaptability or versatility

If forward automatic speech recognition is used on pre-wakeword speech, then processing natural speech patterns is improved, but false positives increase due to non-command speech

Engineering Contradiction:

Improvenatural speech pattern recognitionVSAvoidfalse positive rate

Core Design Contradiction:

Adaptability or versatilityVSReliability

Solution Approach 1:

The patent uses reverse automatic speech recognition to process buffered speech from the wakeword backward in time. This inversion allows the system to naturally accommodate various speech patterns while stopping processing as soon as a complete command is identified, thereby avoiding false positives from non-command speech that would be included in forward processing.

Inventive Principle:

Principle #13The other way round (Inversion)

4Reliability

If the system processes all buffered pre-wakeword speech, then no commands are missed, but processing time increases

Engineering Contradiction:

Improvecommand detection completenessVSAvoidprocessing time

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

By processing speech in reverse from the wakeword backward, the system can identify complete commands more quickly and stop processing early. This eliminates the need to process the entire buffered duration, reducing processing time while maintaining completeness through the fixed buffer that guarantees sufficient coverage of pre-wakeword speech.

Inventive Principle:

Principle #13The other way round (Inversion)

Data Source

PatentUS12424215B2Pre-wakeword speech processing with reverse automatic speech recognition

Publication Date: 2025.09.23 SOUNDHOUND AI IP LLC

AI summary

Methods and systems for pre-wakeword speech processing are disclosed. Speech audio, comprising command speech spoken before a wakeword, may be stored in a buffer in oldest to newest order. Upon detection of the wakeword, reverse acoustic models and language models, such as reverse automatic speech recognition (R-ASR) can be applied to the buffered audio, in newest to oldest order, starting from before the wakeword. The speech is converted into a sequence of words. Natural language grammar models, such as natural language understanding (NLU), can be applied to match the sequence of words to a complete command, the complete command being associated with invoking a computer operation.