Speech emotion recognition (SER) is an active area of research, yet existing work has focused almost exclusively on high-resource languages, leaving Nepali — with over 32 million first- and second-language speakers worldwide — without any published SER study or emotional-speech corpus. This paper addresses that gap along two dimensions. First, we construct a Nepali emotional-speech dataset comprising 600 utterances across three emotion classes (happy, sad, neutral), validated by 117 native listeners whose mean recognition accuracy is 91.5%. Second, on this corpus, we evaluate a fully quantum data-reuploading variational quantum circuit (VQC) classifier with trainable SU(2) encoding on nine qubits, and compare it directly against two classical baselines — a random forest and a multilayer perceptron — on the same PCA(27) feature pipeline and stratified 480/120 split. A staged hyperparameter search covering circuit depth, learning rate, optimizer, and batch size identifies an optimal VQC configuration of ten layers and 540 trainable parameters, which attains 90.83% test accuracy and a macro-F1 of 0.908. Gradient-norm analysis confirms the absence of barren plateaus during training. Both classical baselines outperform the VQC under this protocol (Random Forest 95.00%, MLP 99.17%); however, a leave-one-speaker-out robustness check shows that classical accuracy collapses by approximately one-third under this evaluation, indicating that a substantial portion of the classical advantage reflects speaker-level information leakage.
Speech emotion recognition (SER) is an active area of research, yet existing work has focused almost exclusively on high-resource languages, leaving Nepali — with over 32 million first- and second-language speakers worldwide — without any published SER study or emotional-speech corpus. This paper addresses that gap along two dimensions. First, we construct a Nepali emotional-speech dataset comprising 600 utterances across three emotion classes (happy, sad, neutral), validated by 117 native listeners whose mean recognition accuracy is 91.5%. Second, on this corpus, we evaluate a fully quantum data-reuploading variational quantum circuit (VQC) classifier with trainable SU(2) encoding on nine qubits, and compare it directly against two classical baselines — a random forest and a multilayer perceptron — on the same PCA(27) feature pipeline and stratified 480/120 split. A staged hyperparameter search covering circuit depth, learning rate, optimizer, and batch size identifies an optimal VQC configuration of ten layers and 540 trainable parameters, which attains 90.83% test accuracy and a macro-F1 of 0.908. Gradient-norm analysis confirms the absence of barren plateaus during training. Both classical baselines outperform the VQC under this protocol (Random Forest 95.00%, MLP 99.17%); however, a leave-one-speaker-out robustness check shows that classical accuracy collapses by approximately one-third under this evaluation, indicating that a substantial portion of the classical advantage reflects speaker-level information leakage.
We evaluate a hybrid quantum–classical pipeline for speech emotion recognition (SER) on a custom Afrikaans corpus using MFCC-based spectral features with pitch and energy variants, explicitly comparing three quantum approaches—a variational quantum classifier (VQC), a quantum support vector machine (QSVM), and a Quantum Approximate Optimisation Algorithm (QAOA)-based classifier—against a CNN–LSTM (CLSTM) baseline. We detail the classical-to-quantum data encoding (angle embedding with bounded rotations and an explicit feature-to-qubit map) and report test accuracy, weighted precision, recall, and F1. Under ideal analytic simulation, the quantum models reach 41–43% test accuracy; under a realistic 1% NISQ noise model (100–1000 shots) this degrades to 34–40%, versus 73.9% for the CLSTM baseline. Despite the markedly lower empirical accuracy—expected in the NISQ era—we provide an end-to-end, noise-aware hybrid SER benchmark and discuss the asymptotic advantages of quantum subroutines (Chebyshev-based quantum singular value transformation, quantum walks, and block encoding) that become relevant only in the fault-tolerant regime.
Speech emotion recognition (SER) remains fragile in real-world conditions because emotional cues are subtle, speaker-dependent, and easily confounded by recording variability, while high-performing deep models typically rely on large and carefully curated training sets. Quantum machine learning offers an alternative way to introduce nonlinear correlation modeling with compact modules, yet existing quantum SER studies remain limited, and the impact of circuit structure is not well understood. This paper presents HQTN-SER, a hybrid quantum-classical framework that investigates how quantum tensor network connectivity can support SER under small-qubit settings. HQTN-SER introduces (i) an MPS-inspired quantum tensor network module that enforces structured interactions to model correlations in speech representations with a small number of trainable parameters, and (ii) a fusion strategy that combines quantum measurement features with a learned classical latent embedding for end-to-end emotion classification. We evaluate HQTN-SER on three public benchmarks (RAVDESS, SAVEE, and MDER) under a unified preprocessing and training protocol. The proposed model achieves test accuracies of 80.09\% on RAVDESS, 78.12\% on SAVEE, and 73.51\% on MDER, with corresponding macro-F1 scores of 0.8012, 0.7826, and 0.7351, respectively. The model also exhibits stable convergence while using only three or four qubits, suggesting that MPS-inspired circuit structure can be an effective and hardware-aware design choice for quantum-assisted SER. These results provide a compact experimental baseline for studying structured quantum modules in affective computing.
Multimodal emotion recognition can benefit from combining complementary information from facial images, speech, and text.However, integrating heterogeneous modalities obtained from independently collected datasets introduces challenges in data alignment and label consistency.This study presents a hybrid quantum-classical framework that combines modality-specific classical encoders with a quantum re-uploading variational classifier for six-class emotion recognition.Facial images, speech samples, and textual samples were obtained from separate publicly available datasets and aligned at the emotion-label level to construct a unified multimodal experimental setting.Image, speech, and text encoders generated modality-specific representations that were fused through a classical neural network and projected into a six-dimensional quantum input space.A six-qubit quantum circuit with three re-uploading layers repeatedly encoded the fused features before six-class classification.On a test set of 6,347 samples, the classical Image-Speech-Text fusion model achieved 73.94% accuracy and a 70.64% macro F1-score, whereas the proposed quantum reuploading model achieved 72.57% accuracy and a 68.76% macro F1-score.The quantum model used 8,706 trainable parameters compared with 15,440,006 for the classical multimodal model, representing a 99.94% reduction in trainable parameters.The results indicate that quantum re-uploading can provide competitive performance for multimodal emotion recognition while highlighting the limitations of label-level alignment across independently collected datasets.
Emotion recognition in speech is essential for human-computer interaction. This study introduces a refined hybrid Quantum Convolutional Neural Network (QCNN) framework designed to substantially enhance Speech Emotion Recognition (SER). It addresses major challenges such as speaker variability, noise robustness, and cross-domain applicability that often limit classical deep learning models. By Utilizing quantum principles like superposition and entanglement within a 6-qubit, 8-layer parameterized circuit, the model derives highdimensional, noise-resistant features from speech signals. The method includes sophisticated preprocessing steps, such as bandpass filtering and energy-based Voice Activity Detection (VAD), followed by Mel-Frequency Cepstral Coefficients (MFCC) extraction. Evaluated on a diverse dataset of 4,515 samples across eight emotions (Angry, Calm, Disgust, Fear, Happy, Neutral, Sad, Surprised), the QCNN achieved an accuracy of 86 %, setting a new benchmark. It surpasses classical approaches such as a CNN with attention(77%), a traditional QCNN(77.87%), and a hybrid quantum-classical network(76%). The model demonstrated balanced performance, with precision scores ranging from 0.79 to 1.0 (including a perfect 1.0 for Fear) and recall scores from 0.82 to 0.94. These results underscore the promise of quantum-enhanced neural networks in capturing complex, nonlinear speech emotion patterns and establishing a new standard for robust, generalizable SER systems.
Multimodal emotion recognition can benefit from combining complementary information from facial images, speech, and text. However, integrating heterogeneous modalities obtained from independently collected datasets introduces challenges in data alignment and label consistency. This study presents a hybrid quantum–classical framework that combines modality-specific classical encoders with a quantum re-uploading variational classifier for six-class emotion recognition. Facial images, speech samples, and textual samples were obtained from separate publicly available datasets and aligned at the emotion-label level to construct a unified multimodal experimental setting. Image, speech, and text encoders generated modality-specific representations that were fused through a classical neural network and projected into a six-dimensional quantum input space. A six-qubit quantum circuit with three re-uploading layers repeatedly encoded the fused features before six-class classification. On a test set of 6,347 samples, the classical Image–Speech–Text fusion model achieved 73.94% accuracy and a 70.64% macro F1-score, whereas the proposed quantum re-uploading model achieved 72.57% accuracy and a 68.76% macro F1-score. The quantum model used 8,706 trainable parameters compared with 15,440,006 for the classical multimodal model, representing a 99.94% reduction in trainable parameters. The results indicate that quantum re-uploading can provide competitive performance for multimodal emotion recognition while highlighting the limitations of label-level alignment across independently collected datasets.
Quantum machine learning (QML) offers a promising avenue for advancing representation learning in complex signal domains. In this study, we investigate the use of parameterised quantum circuits (PQCs) for speech emotion recognition (SER)-a challenging task due to the subtle temporal variations and overlapping affective states in vocal signals. We propose a hybrid quantum-classical architecture that integrates PQCs into a conventional convolutional neural network (CNN), leveraging quantum properties such as superposition and entanglement to enrich emotional feature representations. Experimental evaluations on three benchmark datasets IEMOCAP, RECOLA, and MSP-IMPROV-demonstrate that our hybrid model achieves improved classification performance relative to a purely classical CNN baseline, with over 50% reduction in trainable parameters. Furthermore, Adjusted Rand Index (ARI) analysis demonstrates that the quantum model yields feature representations with improved alignment to true emotion classes compared with the classical model, reinforcing the observed performance gains. This work provides early evidence of the potential for QML to enhance emotion recognition and lays the foundation for future quantum-enabled affective computing systems.