Towards Reliable DTMF Recognition: A TinyML Approach

Dual-tone multi-frequency (DTMF) detection faces challenges in real-world environments due to noise, distortions, and diverse telecommunications equipment. This paper explores the application of a simple Artificial Neural Network (ANN) for robust DTMF tone classification, with subsequent deployment on an Arduino device. The proposed TinyML model achieves near-identical accuracy to the unquantized model across pure signals, signals with noise, and speech distortions. Notably, the quantized model significantly reduces file size (70%) while maintaining excellent performance. This work highlights the potential of ANNs for robust DTMF classification in real-world scenarios, particularly on low-power devices.