Skip to content

SPEAR-ASR and SPEAR-WakeUp Software Development Kit in Java for Linux

License

Notifications You must be signed in to change notification settings

Think-A-Move/SPEAR-SDK-Java-Linux

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SPEAR-SDK-Java-Linux

SPEAR-SDK-Java-Linux is SPEAR's software development kit that allows you to integrate SPEAR-ASR and SPEAR-WakeUp to voice-enable your Java applications in Linux. The SPEAR-ASR engine will terminate after 10 minutes of processing speech. SPEAR-WakeUp will not terminate after 10 minutes. For continuous operation, contact us at [email protected].

By cloning this repository and downloading SPEAR-SDK library you agree to the SPEAR-SDK EULA.

Contents:

  • SpearSdk-5.1.0.jar - SPEAR-SDK library for Java consisting of SPEAR-ASR and SPEAR-WakeUp.

  • SPEAR-ASR is an on-device automatic speech recognition engine that delivers superior recognition accuracy in high noise environments and provides both command recognition and free-form transcription. Developed for use by military medics in the battlefield.

  • SPEAR-WakeUp is an on-device lightweight Wake word detection system that has very low power consumption.

  • Documentation for the API and app integration is located in the docs directory.

  • SpearSdkExample is an example JavaFx app that demonstrates how to integrate our SPEAR-SDK library and language model with JavaFx applications

Features

  • Noise-robust DNN-Based Acoustic Model
  • Noise-robust DNN-Based VAD Model
  • Domain-specific Language Model
  • Works Offline
  • Supports NBest results
  • Supports Command and Control (finite state) Applications
  • Supports Transcriptions (statistical)
  • Speaker Independent
  • Grammar switching at runtime
  • Dynamic grammar compilation (Grammar on-the-fly)
  • Supports 16K Audio Sampling Rate
  • Down-sample to 8K or 16K
  • Low CPU consumption of Wake-up Word
  • Java language support
  • Floating point version
  • Barge-in support

Languages

  • Chinese - Mandarin
  • English
  • French
  • Spanish
  • We can train other languages or augment existing languages. Contact us at [email protected].

Acoustic Models

  • Optimized for high-noise
  • We can train acoustic models for specific noise environments. Contact us at [email protected].
  • Accuracy results using WSJ Corpus mixed with various noise samples WSJ Accuracy Results

Services we can provide

  • Create customized Acoustic Models
  • Create domain Specific Language Models
  • Extend additional functionality of SPEAR-ASR to address specific needs
  • Voice-enable your application or make one for you
  • Contact us at [email protected].