SEEF-ALDR: A Speaker Embedding Enhancement Framework via Adversarial Learning based Disentangled Representation

Tai, Jung-Hsiang; Jia, Xiaoqi; Huang, Qilin; Zhang, Weijuan; Du, Haichao; Zhang, Shengzhi

doi:10.1145/3427228.3427274

Cited by 3 publications

(4 citation statements)

References 51 publications

(109 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…All models for comparison are re-implemented by ours. Table 1 shows the equal error rate (EER) obtained by the Vox-Celeb1 [8] testset, where we compare our models with the encoder model [16] and the disentanglement model [5]. With the standard softmax loss and TAP aggregation, our model outperforms previous models based on the ResNet encoder by 36.6% and the disentanglement model using an adversarial method [5] by 16.9%.…”

Section: Resultsmentioning

confidence: 97%

“…Table 1 shows the equal error rate (EER) obtained by the Vox-Celeb1 [8] testset, where we compare our models with the encoder model [16] and the disentanglement model [5]. With the standard softmax loss and TAP aggregation, our model outperforms previous models based on the ResNet encoder by 36.6% and the disentanglement model using an adversarial method [5] by 16.9%. These results demonstrate that the represented embeddings of the proposed disentanglement approach are more informative than those of the baseline.…”

Section: Resultsmentioning

confidence: 97%

“…Tai at el. [5] proposed a disentanglement method for speaker recognition that is the baseline for our work. By constructing an identity-related and an identity-unrelated encoder, they trained each encoder to represent only speaker-related and -unrelated information using speaker identification loss and adversarial training loss.…”

Section: Disentangled Feature Learningmentioning

confidence: 99%

“…To overcome the aforementioned limitation inherent to the encoder style framework, a method for disentangling the embeddings with the use of relevant and irrelevant speaker information was proposed [5]. The method consists of two encoders, a speaker purifying encoder and a dispersing encoder, as well as a decoder for reconstruction.…”

Section: Introductionmentioning

confidence: 99%

See 3 more Smart Citations

Intra-Class Variation Reduction of Speaker Representation in Disentanglement Framework

Kwon¹,

Chung²,

Kang³

2020

Interspeech 2020

View full text Add to dashboard Cite

In this paper, we propose an effective training strategy to extract robust speaker representations from a speech signal. One of the key challenges in speaker recognition tasks is to learn latent representations or embeddings containing solely speaker characteristic information in order to be robust in terms of intraspeaker variations. By modifying the network architecture to generate both speaker-related and speaker-unrelated representations, we exploit a learning criterion which minimizes the mutual information between these disentangled embeddings. We also introduce an identity change loss criterion which utilizes a reconstruction error to different utterances spoken by the same speaker. Since the proposed criteria reduce the variation of speaker characteristics caused by changes in background environment or spoken content, the resulting embeddings of each speaker become more consistent. The effectiveness of the proposed method is demonstrated through two tasks; disentanglement performance, and improvement of speaker recognition accuracy compared to the baseline model on a benchmark dataset, VoxCeleb1. Ablation studies also show the impact of each criterion on overall performance.

show abstract