Tone classification of online medical services based on 1DCNN-BiLSTM

Area of Science:

• Artificial Intelligence
• Speech Processing
• Medical Informatics

Background:

• Accurate tone classification in online medical services is crucial for effective doctor-patient communication.
• Existing methods may not fully capture the complex acoustic features of doctor's speech in this context.

Purpose of the Study:

• To develop and evaluate a novel model for enhancing the recognition rate of doctor tone classification in online medical scenarios.
• To improve the efficiency and quality of doctor-patient communication through better tone analysis.

Main Methods:

• A hybrid model integrating a one-dimensional convolutional neural network (1DCNN) for local feature extraction and a bidirectional long short-term memory (BiLSTM) network for global sequential feature capture was proposed.
• A survey identified significant tone types, and 68 time- and frequency-domain features were extracted using Librosa.
• Feature-level fusion combined the strengths of both 1DCNN and BiLSTM.

Main Results:

• The proposed 1DCNN-BiLSTM model achieved an average recognition rate of 84.4% and an F1 score of 84.4% in online medical service scenarios.
• The model significantly outperformed existing methods in tone classification accuracy.
• Ablation experiments confirmed the effectiveness of individual modules (1DCNN, BiLSTM) and parameter settings.

Conclusions:

• The integrated 1DCNN-BiLSTM model offers a significant improvement in doctor tone classification accuracy within online medical settings.
• This approach effectively enhances doctor-patient communication by providing better insights into vocal cues.
• The study validates the utility of hybrid deep learning architectures for complex audio analysis in specialized domains.