Tactical Employment of Audio with Mobile Machine Learning (TEAM-ML)

The objective of this effort is delivering cutting-edge audio-based machine learning models that effectively leverage TAK-ML architecture for seamless in-field execution in a mobile environment. At project conclusion, we will deliver machine learning models optimized for TAK-ML deployment on tactical edge devices.á Simply running MISTK models on devices is insufficient - they must be tailored for resource-constrained edge execution. The proposed Tactical Employment of Audio with Mobile Machine Learning (TEAM-ML) solution, is a combination of the Unhelpful Node and Dependency Elimination and Reduction to Support Tactical Applications and Networked Deployments (UNDERSTAND) and three (3) machine learning models. UNDERSTANDáwill provide tools to:á Easily extract models from their MISTK wrappersá Identify the Python dependencies Recommend mitigationsá Reduce size and resource requirementsá Compare the results of the reduced vs original modelsá The machine learning models include:á TAK Transcription and Automated Language Kit (TAK TALK) - Real-time voice communications with embedded translation servicesá Distributed Acoustic Reasoner for TAK (DART) - Gunshot localization and tracking via sensor networksá Aircraft Classification Enabled by Machine Learning (ACE-ML) - Aircraft type and threat classification from a distanceá Rigorous testing will occur in operationally relevant environments prior to full deployment. All data and software will be delivered with unlimited data rights, aligning with TAK principles as a government open-source model.á