On Calibration of Speech Classification Models: Insights from Energy-Based Model Investigations
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866910502351798272 |
|---|---|
| author | Hao, Yaqian Hu, Chenguang Gao, Yingying Zhang, Shilei Feng, Junlan |
| author_facet | Hao, Yaqian Hu, Chenguang Gao, Yingying Zhang, Shilei Feng, Junlan |
| contents | For speech classification tasks, deep learning models often achieve high accuracy but exhibit shortcomings in calibration, manifesting as classifiers exhibiting overconfidence. The significance of calibration lies in its critical role in guaranteeing the reliability of decision-making within deep learning systems. This study explores the effectiveness of Energy-Based Models in calibrating confidence for speech classification tasks by training a joint EBM integrating a discriminative and a generative model, thereby enhancing the classifiers calibration and mitigating overconfidence. Experimental evaluations conducted on three speech classification tasks specifically: age, emotion, and language recognition. Our findings highlight the competitive performance of EBMs in calibrating the speech classification models. This research emphasizes the potential of EBMs in speech classification tasks, demonstrating their ability to enhance calibration without sacrificing accuracy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_18065 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | On Calibration of Speech Classification Models: Insights from Energy-Based Model Investigations Hao, Yaqian Hu, Chenguang Gao, Yingying Zhang, Shilei Feng, Junlan Audio and Speech Processing Sound For speech classification tasks, deep learning models often achieve high accuracy but exhibit shortcomings in calibration, manifesting as classifiers exhibiting overconfidence. The significance of calibration lies in its critical role in guaranteeing the reliability of decision-making within deep learning systems. This study explores the effectiveness of Energy-Based Models in calibrating confidence for speech classification tasks by training a joint EBM integrating a discriminative and a generative model, thereby enhancing the classifiers calibration and mitigating overconfidence. Experimental evaluations conducted on three speech classification tasks specifically: age, emotion, and language recognition. Our findings highlight the competitive performance of EBMs in calibrating the speech classification models. This research emphasizes the potential of EBMs in speech classification tasks, demonstrating their ability to enhance calibration without sacrificing accuracy. |
| title | On Calibration of Speech Classification Models: Insights from Energy-Based Model Investigations |
| topic | Audio and Speech Processing Sound |
| url | https://arxiv.org/abs/2406.18065 |