CODEC Keyword Detection Buffering for Speech Recognition Latency

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Mobile devices face delays in processing speech recognition commands when transitioning from a low power state, requiring users to wait for a prompt before providing input, which is inconvenient and inefficient.

Innovation Solution

Implementing a coder/decoder (CODEC) with a keyword detector and a first buffer to buffer audio data until the application processor is ready, allowing for immediate processing of speech recognition commands without the need for a prompt, by transferring audio data to a second buffer once the processor is initialized.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If the mobile device waits for the application processor to wake up and provide a prompt before processing speech recognition, then the processing can be completed accurately, but the user experience is degraded due to delay

Engineering Contradiction:

Improvespeech recognition processing accuracyVSAvoiddelay between user speech and command execution

Core Design Contradiction:

ReliabilityVSLoss of time

Solution Approach 1:

The keyword detector performs preliminary action by detecting keywords in audio data before the application processor is fully initialized. The audio data is buffered in advance, so when the application processor wakes up, the data is already ready for immediate processing, eliminating the need to wait for a user prompt.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

A buffer acts as an intermediary between the audio capture system and the application processor. The buffer stores audio data temporarily, allowing the keyword detector to operate independently from the application processor's power state, and ensuring data is available when the processor is ready.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of information

If a large buffer is used to store audio data during processor initialization, then all audio data can be captured, but the device cost increases

Engineering Contradiction:

Improveaudio data capture completenessVSAvoiddevice cost

Core Design Contradiction:

Loss of informationVSEase of manufacture

Solution Approach 1:

Instead of using a large buffer to capture the entire possible audio stream, the system uses a smaller buffer that captures only the critical audio data during the keyword detection phase. This partial action approach reduces buffer size requirements while maintaining sufficient data capture for speech recognition.

Inventive Principle:

Principle #16Partial or excessive action

Solution Approach 2:

The keyword detector identifies and buffers only the relevant audio segments containing keywords before the processor initializes. This preliminary filtering action reduces the amount of data that needs to be stored, allowing for a smaller, more cost-effective buffer.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS9652017B2System and method of analyzing audio data samples associated with speech recognition

Publication Date: 2017.05.16 QUALCOMM INC

AI summary

A particular apparatus includes a coder/decoder (CODEC) including a first processor and a first buffer. The first processor is configured to analyze audio data samples to detect a keyword. The CODEC is configured to store a set of audio data samples at the first buffer. The apparatus also includes an application processor configured to receive the set of audio data samples from the CODEC via a bus and configured to initialize a speech recognition engine (SRE) based on the set of audio data samples. The application processor is configured to initialize the bus based on an indication from the CODEC that the keyword is detected.