Demonstration of Adapt4Me: An Uncertainty-Aware Authoring Environment for Personalizing Automatic Speech Recognition to Non-normative Speech

1. 1Data collection is crowdsourced through the web-based interface
2. 2Bayesian active learning is applied to select the most informative data samples
3. 3Model adaptation and validation are performed without expert supervision
4. 4End-users can validate and refine the personalized ASR model

Speech recognition engineers and researchers can benefit from Adapt4Me as it simplifies the personalization process, while product managers can leverage it to improve user experience for individuals with non-normative speech

💡 Bayesian active learning enables efficient and accurate personalization of ASR models without requiring large amounts of labeled data or expert supervision

🗣️ Adapt4Me personalizes ASR for non-normative speech using Bayesian active learning! 🚀