Audio Applications in the TinyML Era

This one-hour talk explores how tinyML technology is revolutionizing audio applications on microcontroller units (MCUs) with limited computational resources. Discover two transformative use cases: single-channel environmental noise cancellation (ENC) for improved communication and the innovative text2model (T2M) approach for open vocabulary keyword spotting that enables custom command definition without extensive data requirements. Learn practical implementation strategies including quantization-aware training, SVD compression, and distillation techniques, while addressing real-world challenges like PyTorch to TensorFlow Lite conversion and CNN streaming. Gain valuable insights into deploying sophisticated audio AI applications on resource-constrained devices, bridging the gap between research concepts and practical implementation.