On-Device Speech Service Interface for Multilingual Low-Latency Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing speech recognition technologies lack efficient on-device multilingual capabilities and latency reduction, particularly in cloud-free environments, which hinders seamless integration and real-time streaming capabilities.

Innovation Solution

A client-side speech service interface that integrates multilingual speech recognition, enabling dynamic language switching and speaker detection through configuration parameters, utilizing language ID predictor models and candidate language packs, and buffering mechanisms to enhance accuracy and latency reduction.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If cloud-based speech recognition is used, then computing resources are abundant, but latency increases and user privacy is compromised

Engineering Contradiction:
Improvespeech recognition capabilityVSAvoidlatency
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent extracts the speech recognition functionality from the cloud and implements it on-device using a speech service interface that includes a speech recognition engine, language ID predictor model, and multiple language packs stored locally on the client device. This extraction eliminates network dependency and reduces latency while maintaining recognition capability.

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent implements preliminary action by pre-loading and storing multiple language packs (e.g., English, Spanish, French) and the speech recognition engine on the client device before actual speech recognition is needed. The language pack directory structure is pre-configured with mappings between language codes and local file paths, enabling immediate execution without cloud connection or model download during speech processing.

Inventive Principle:
Principle #10Preliminary action

2Adaptability or versatility

If multiple language packs are loaded onto the device, then multilingual capability is improved, but device memory and processing resources are consumed

Engineering Contradiction:
Improvemultilingual speech recognition capabilityVSAvoidstorage and processing requirements
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent implements dynamics by creating a selective loading mechanism where the speech service interface dynamically loads only the required language pack based on the detected language code. The language pack directory contains multiple language packs (e.g., en-US, es-ES, fr-FR) but the system loads only the active one into memory, making the system adaptable to multiple languages without permanently consuming resources for all languages simultaneously.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent segments the speech recognition system into independent language packs, each containing language-specific models and resources. The language pack directory structure organizes these segments separately, allowing the system to load, unload, and switch between segments independently. This segmentation reduces the memory footprint compared to loading all language packs simultaneously while maintaining multilingual capability.

Inventive Principle:
Principle #1Segmentation

3Productivity

If real-time streaming speech recognition is implemented, then user experience is improved, but accuracy may be compromised with limited computational resources

Engineering Contradiction:
Improvestreaming capabilityVSAvoidspeech recognition accuracy
Core Design Contradiction:
ProductivityVSMeasurement precision

Solution Approach 1:

The patent ensures continuity of useful action by implementing a streaming architecture where the speech service interface continuously processes audio data in real-time. The system maintains an active state with the speech recognition engine continuously analyzing audio streams, language ID predictor models continuously detecting language switches, and transcriptions being emitted continuously as speech is spoken, rather than processing in batches or requiring completion of entire utterances.

Inventive Principle:
Principle #20Continuity of useful action

Solution Approach 2:

The patent applies parameter changes by dynamically adjusting the speech recognition engine's operating parameters based on detected language switches. When the language ID predictor model detects a language change, the system changes the active language pack parameter and reconfigures the recognition engine accordingly. This dynamic parameter adjustment enables the system to maintain high accuracy across multiple languages while preserving real-time streaming performance.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentUS20250378286A1Application Programming Interfaces For On-Device Speech Services
Publication Date: 2025.12.11 GOOGLE LLC
  • US20250378286A1 patent drawing
  • US20250378286A1 patent drawing
  • US20250378286A1 patent drawing

AI summary

A method (500) includes receiving, from an application (50) executing on a client device (110), at a speech service interface (200), configuration parameters (211) for integrating a speech service (250) into the application. The configuration parameters include a language pack directory (225) that maps a primary language code (235) to an on-device path of a primary language pack (110) of the speech service for use in recognizing speech in a primary language and each of one or more codeswitch language codes to an on-device path. The method also includes receiving audio data (102) characterizing an utterance (106) and processing, using a language ID predictor model (230), the audio data to determine that the audio data is associated with the primary language code. The method also includes processing, using the primary language pack, the audio data to determine a transcription (120) that includes one or more words in the primary language.