We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:

Download:

Current browse context:

cs.SD

Change to browse by:

References & Citations

DBLP - CS Bibliography

Bookmark

(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Sound

Title: Binary Neural Network for Speaker Verification

Abstract: Although deep neural networks are successful for many tasks in the speech domain, the high computational and memory costs of deep neural networks make it difficult to directly deploy highperformance Neural Network systems on low-resource embedded devices. There are several mechanisms to reduce the size of the neural networks i.e. parameter pruning, parameter quantization, etc. This paper focuses on how to apply binary neural networks to the task of speaker verification. The proposed binarization of training parameters can largely maintain the performance while significantly reducing storage space requirements and computational costs. Experiment results show that, after binarizing the Convolutional Neural Network, the ResNet34-based network achieves an EER of around 5% on the Voxceleb1 testing dataset and even outperforms the traditional real number network on the text-dependent dataset: Xiaole while having a 32x memory saving.
Subjects: Sound (cs.SD); Machine Learning (cs.LG); Audio and Speech Processing (eess.AS)
Cite as: arXiv:2104.02306 [cs.SD]
  (or arXiv:2104.02306v1 [cs.SD] for this version)

Submission history

From: Tinglong Zhu [view email]
[v1] Tue, 6 Apr 2021 06:04:57 GMT (832kb,D)

Link back to: arXiv, form interface, contact.