RawNet: Advanced End-to-End Deep Neural Network Using Raw Waveforms for Text-Independent Speaker Verification
RawNet: усовершенствованная сквозная глубокая нейронная сеть, использующая исходные речевые сигналы для текстонезависимой верификации говорящего
2019-09-13
SCID: 54.1/uxh2urrx
Discuss with AI
End-to-end deep neural networksRaw waveform modelingSpeaker embeddingsSpeaker verificationVoxCeleb1
Figures from the paper
Abstract (AI)
Recently, direct modeling of raw waveforms using deep neural networks has been widely studied for a number of tasks in audio domains. In speaker verification, however, utilization of raw waveforms is in its preliminary phase, requiring further investigation. In this study, we explore end-to-end deep neural networks that input raw waveforms to improve various aspects: front-end speaker embedding extraction including model architecture, pre-training scheme, additional objective functions, and back-end classification. Adjustment of model architecture using a pre-training scheme can extract speaker embeddings, giving a significant improvement in performance. Additional objective functions simplify the process of extracting speaker embeddings by merging conventional two-phase processes: extracting utterance-level features such as i-vectors or x-vectors and the feature enhancement phase, e.g., linear discriminant analysis. Effective back-end classification models that suit the proposed speaker embedding are also explored. We propose an end-to-end system that comprises two deep neural networks, one front-end for utterance-level speaker embedding extraction and the other for back-end classification. Experiments conducted on the VoxCeleb1 dataset demonstrate that the proposed model achieves state-of-the-art performance among systems without data augmentation. The proposed system is also comparable to the state-of-the-art x-vector system that adopts data augmentation.
Key Findings
1
Additional objective functions integrate speaker embedding extraction and feature enhancement, replacing conventional two-stage i-vector or x-vector pipelines with a unified process.
2
Architecture adjustment combined with pre-training substantially improves raw-waveform speaker embedding extraction.
3
On VoxCeleb1, RawNet achieves state-of-the-art performance among systems without data augmentation and performs comparably to an augmented x-vector system.
4
RawNet directly models raw waveforms with an end-to-end deep neural network for text-independent speaker verification.
5
The proposed system uses separate front-end and back-end neural networks for utterance-level speaker embedding extraction and speaker classification.
Research Object
end-to-end deep neural network systems using raw waveforms for text-independent speaker verification
Research Subject
speaker embedding extraction and back-end classification performance, including the effects of model architecture, pre-training, additional objective functions, and classification models
Publication Details
Publication Date
2019-09-13
Journal
Publisher
ISSN
Cited by
166
Open access PDF
Access Type
Author Information
Download PDF
Subscribe to digest
References available in scid.ai5
Learning Phrase Representations using RNN Encoder–Decoder for Statistical Machine Translation2014
Empirical Evaluation of Gated Recurrent Neural Networks on Sequence Modeling2014
Front-End Factor Analysis for Speaker Verification2010
X-Vectors: Robust DNN Embeddings for Speaker Recognition2018
VoxCeleb: A Large-Scale Speaker Identification Dataset2017