This study proposes a comprehensive and computationally efficient system for the recognition of heart valve diseases (HVDs) in phonocardiogram (PCG) signals, emphasizing an end-to-end workflow suitable for real-world deployment. The core of the system is a lightweight weighted convolutional neural network (WCNN) featuring a key weighting calculation (KWC) layer, which enhances noise robustness by adaptively weighting feature map channels based on global average pooling. The proposed system incorporates optimized feature extraction using Mel-frequency cepstral coefficients (MFCCs) guided by GradCAM, and a band energy ratio (BER) metric to assess signal quality, showing that lower BER values are associated with higher misclassification rates due to noise. Experimental results demonstrated classification accuracies of 99.6% and 90.74% on the GitHub PCG and PhysioNet/CinC Challenge 2016 databases, respectively, where the models were trained and tested independently. The proposed model achieved superior accuracy using significantly fewer parameters (312,357) and lower computational cost (4.5 M FLOPs) compared with previously published research. Compared with the model proposed by Karhade et al., the proposed model use 74.9% fewer parameters and 99.3% fewer FLOPs. Furthermore, the proposed model was implemented on a Raspberry Pi, achieving real-time HVDs detection with a detection time of only 1.87 ms for a 1.4 s signal.

As voice cloning technology rapidly advances, the risk of personal voices being misused by malicious actors for fraud or other illegal activities has significantly increased, making the collection of speech data increasingly challenging. To address this issue, this study proposes a data augmentation method based on XText-to-Speech (XTTS) synthesis to tackle the challenges of small-sample, multi-class speech recognition, using profanity as a case study to achieve high-accuracy keyword recognition. Two models were therefore evaluated: a CNN model (Proposed-I) and a CNN-Transformer hybrid model (Proposed-II). Proposed-I leverages local feature extraction, improving accuracy on a real human speech (RHS) test set from 55.35% without augmentation to 80.36% with XTTS-enhanced data. Proposed-II integrates CNN’s local feature extraction with Transformer’s long-range dependency modeling, further boosting test set accuracy to 88.90% while reducing the parameter count by approximately 41%, significantly enhancing computational efficiency. Compared to a previously proposed incremental architecture, the Proposed-II model achieves an 8.49% higher accuracy while reducing parameters by about 98.81% and MACs by about 98.97%, demonstrating exceptional resource efficiency. By utilizing XTTS and public corpora to generate a novel keyword speech dataset, this study enhances sample diversity and reduces reliance on large-scale original speech data. Experimental analysis reveals that an optimal synthetic-to-real speech ratio of 1:5 significantly improves the overall system accuracy, effectively addressing data scarcity. Additionally, the Proposed-I and Proposed-II models achieve accuracies of 97.54% and 98.66%, respectively, in distinguishing real from synthetic speech, demonstrating their strong potential for speech security and anti-spoofing applications.