Confidence Learning for Semi-Supervised Acoustic Event Detection

In recent years, the involvement of synthetic strongly labeled data, weakly labeled data, and unlabeled data has drawn much research attention in semi-supervised acoustic event detection (SAED). The classic self-training method carries out predictions for unlabeled data and then selects predictions with high probabilities as pseudo-labels for retraining. Such models have shown its effectiveness in SAED. However, probabilities are poorly calibrated confidence estimates, and samples with low probabilities are ignored. Hence, we introduce a confidence-based semi-supervised Acoustic event detection (C-SAED) framework. The C-SAED method learns confidence deliberately and retrains all data distinctly by applying confidence as weights. Additionally, we apply a power pooling function whose coefficient can be trained automatically and use weakly labeled data more efficiently. The experimental results demonstrate that the generated confidence is proportional to the accuracy of the predictions. Our C-SAED framework achieves a relative error rate reduction of 34% in contrast to the baseline model.

Passive Extraction of Signal Feature Using a Rectifier with a Mechanically Switched Inductor for Low Power Acoustic Event Detection

Analog hardware used for signal envelope extraction in low-power interfaces for acoustic event detection, owing to its low complexity and power consumption, suffers from low sensitivity and performs poorly under low signal to noise ratios (SNR) found in undersea environments. To overcome those problems, in this paper, we propose a novel passive electromechanical solution for the signal feature extraction in low frequency acoustic range (200–1000 Hz), in the form of a piezoelectric vibration transducer, and a rectifier with a mechanically switched inductor. A simulation study of the novel solution is presented, and a proof-of-concept device is developed and experimentally characterized. We demonstrate its applicability and show the advantages of the passive electromechanical device in comparison to the active electrical solution in terms of operation with lower input signals (<20 mV compared to 40 mV), and higher robustness in low SNR conditions (output voltage loss for −10 dB ≤ SNR < 40 dB of 1 mV, compared to 10 mV). In addition to the signal processing performance improvements, compared to our previous work, the utilization of the presented novel passive feature extractor would also decrease power consumption of a detector’s channel by over 76%, enabling life-time extension and/or increased quality of detection with larger number of channels. To the best of our knowledge, this is the first solution presented in the literature that demonstrates the possibility of using a passive electromechanical feature extractor in a low-power analog wake-up event detector interface.