Distributed Speech Recognition via Parallel Stream Segmentation

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing speech recognition systems face challenges in achieving fast response times without overburdening limited computing resources on client devices, and conventional server-side speech recognition techniques are hindered by unreliable network connections and communication restrictions.

Innovation Solution

A speech recognition client sends a speech stream and control stream in parallel to a server-side recognizer over a network, allowing the server to continuously recognize speech and reconfigure as needed, using a connectionless protocol to minimize delays and optimize resource usage.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of time

If speech recognition is performed on the end user device, then response time is reduced, but computing resources are consumed

Engineering Contradiction:

Improveresponse timeVSAvoidcomputing resources

Core Design Contradiction:

Loss of timeVSUse of energy by moving object

Solution Approach 1:

The system divides speech recognition functionality into segments: resource-intensive processing (speech models, recognition engine) is separated and placed on a server, while lightweight client components handle local operations. This segmentation allows fast local response for simple tasks while maintaining accurate server-side recognition for complex processing.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

A server acts as an intermediary between the end user device and speech recognition resources. The server hosts the full speech recognition system and communicates with client devices, providing recognition capabilities without requiring full installation on each device. This intermediary approach balances response time and resource consumption.

Inventive Principle:

Principle #24Intermediary (Mediator)

2Loss of time

If speech recognition system is installed on every end user device, then response time is reduced, but installation and maintenance complexity increases

Engineering Contradiction:

Improveresponse timeVSAvoidinstallation and maintenance

Core Design Contradiction:

Loss of timeVSDevice complexity

Solution Approach 1:

The complex speech recognition system is extracted from individual end user devices and centralized on a server. Only lightweight client components remain on devices, eliminating the need for full installation and maintenance on each device while preserving fast local response capabilities.

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

A universal server-based speech recognition system serves multiple end user devices through a network. This single centralized system provides recognition capabilities to all clients, eliminating the need for separate installations on each device while maintaining consistent performance across the network.

Inventive Principle:

Principle #6Universality (Multi-functionality)

3Loss of time

If speech recognition is performed on handheld mobile devices, then response time is reduced, but recognition accuracy decreases

Engineering Contradiction:

Improveresponse timeVSAvoidrecognition accuracy

Core Design Contradiction:

Loss of timeVSMeasurement precision

Solution Approach 1:

Recognition functionality is segmented between device and server: simple, fast recognition tasks are handled locally on the mobile device for immediate response, while complex or ambiguous speech segments are transmitted to the server for accurate processing. This segmentation maintains both fast response time and high recognition accuracy.

Inventive Principle:

Principle #1Segmentation

Solution Approach 2:

The server acts as an intermediary that enhances the limited mobile device recognition capabilities. The device handles initial processing for speed, then consults the server for accurate recognition when needed, combining the advantages of both local fast processing and server-based accurate recognition.

Inventive Principle:

Principle #24Intermediary (Mediator)

4Use of energy by moving object

If server-side speech recognition is used, then computing resources are reduced, but network reliability requirements increase

Engineering Contradiction:

Improvecomputing resourcesVSAvoidnetwork connection

Core Design Contradiction:

Use of energy by moving objectVSReliability

Solution Approach 1:

Speech segments are pre-processed and buffered on the client device before transmission to the server. This preliminary action ensures that even if network connections are unreliable or intermittent, the server receives complete speech data for accurate recognition, reducing the impact of network issues on overall system reliability.

Inventive Principle:

Principle #10Preliminary action

Data Source

PatentUS9502033B2Distributed speech recognition using one way communication

Publication Date: 2016.11.22 SOLVENTUM INTELLECTUAL PROPERTIES CO

AI summary

A speech recognition client sends a speech stream and control stream in parallel to a server-side speech recognizer over a network. The network may be an unreliable, low-latency network. The server-side speech recognizer recognizes the speech stream continuously. The speech recognition client receives recognition results from the server-side recognizer in response to requests from the client. The client may remotely reconfigure the state of the server-side recognizer during recognition.