The latest devAIce® SDK 3.14 and Web API 4.7 releases bring a significant boost to the Expression (Large) module, making it more accurate and efficient than ever before. 
 

Upgrades for enhanced accuracy & performance 

The Expression (Large) module’s dimensional and categorical models have been updated with advanced versions that improve recognition precision while reducing resource consumption. The new model has been trained on more data, including new languages, further improving its robustness.  
Notably, categorical output accuracy has been significantly enhanced, ensuring more reliable expression analysis. The Unweighted Average Recall (UAR) has been improved from 0.65 to 0.70 when evaluated on multiple test sets consisting of both acted and non-acted expressions, featuring a multitude of different speakers, languages, microphones, and acoustic environments. 
Additionally, when both categorical and dimensional outputs are enabled, the module now operates more efficiently, optimizing overall performance. 

Additional improvements for devAIce® SDK

This update also includes fixes and enhancements across the SDK, including ASR language detection corrections and improved RT₆₀ output in the Audio Quality module. 
The documentation is also improved and extended with additional information about SDK parallel usage and containerization, which seems to be a common way of how the devAIce® SDK is used. 
 

Your takeaways from the latest  devAIce® upgrades 

Summarizing the major improvements, you get with devAIce® SDK 3.14 and Web API 4.7 that you should know: 

• Improved Expression (Large) Accuracy: Categorical output UAR increased from 0.65 to 0.70, delivering more precise expression analysis across diverse environments. 

• Optimized Performance: More efficient processing when using both categorical and dimensional outputs, reducing resource consumption from 900 MB to 550 MB. 

• Expanded Language Support: Training on additional data and languages for greater robustness. 

• Enhanced ASR & Audio Quality: Fixes in ASR language detection and improved RT₆₀ output in the Audio Quality module. 

• Better Documentation: Extended information on parallel SDK usage and containerization to support real-world deployment needs. 
   
  Upgrade now to take advantage of these powerful improvements! Contact us if you want to start your voice-journey: sales@audeering.com