Phonetic variability constrained bottleneck features for joint speaker recognition and physical task stress...

Area of Science:

• Speech processing
• Machine learning
• Biometrics

Background:

• Robust speech and speaker recognition systems require normalization of intrinsic variabilities like aging and task stress.
• Physical task stress significantly impacts speech production, posing challenges for recognition models.

Purpose of the Study:

• To propose an innovative framework using deep neural networks (DNNs) for joint text-independent speaker recognition and physical task stress detection.
• To analyze speech under physical task stress for improved system robustness.

Main Methods:

• A DNN-based framework utilizing phonetic variability constrained feature vectors as input.
• Employing a universal background model (UBM) for acoustic feature alignment.
• Generating innovative feature representations by selecting and concatenating frames based on UBM alignments.

Main Results:

• The proposed method constrains phonetic variability in feature vectors, alleviating data imbalance and overfitting.
• Experiments on the UTScope-Physical Task Stress Corpus show improved accuracy and Equal Error Rate.
• Performance surpasses a strong i-vector probabilistic linear discriminant analysis system.

Conclusions:

• The proposed DNN framework effectively normalizes intrinsic speech variabilities for enhanced speaker recognition and stress detection.
• The novel feature representation method demonstrates significant improvements in system robustness and performance.