In defence of metric learning for speaker recognition

Chung, Joon Son; Huh, Jaesung; Mun, Seongkyu; Lee, Minjae; Heo, Hee Soo; Choe, Soyeon; Ham, Chiheon; Jung, Sunghwan; Lee, Bong-Jin; Han, Icksang

doi:10.48550/arxiv.2003.11982

Cited by 33 publications

(71 citation statements)

References 39 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We denote the proposed model as STB-ASV. For the STB-ASV training, the network structure of its initial singlechannel ASV is the same as in [23], which contains three main components: a front-end residual convolution neural network (ResNet) [18], a self-attentive pooling (SAP) [24] layer and a fully-coonnected layer. It was trained for 200 epochs on the Librispeech corpus.…”

Section: Methodsmentioning

confidence: 99%

Multi-Channel Far-Field Speaker Verification with Large-Scale Ad-hoc Microphone Arrays

Liang¹,

Chen²,

Yao³

et al. 2021

Preprint

View full text Add to dashboard Cite

Ad-hoc microphone arrays has recieved attention, in which the number and arrangement of microphones are unknown. Traditional multi-channel processing methods can not be directly used in ad-hoc. Recently, to solve this problem, an utterance-level ASV with ad-hoc microphone arrays has been proposed, which first extracts utterance-level speaker embeddings from each channel of an ad-hoc microphone array, and then fuses the embeddings for the final verification. However, this method cannot make full use of the cross-channel information. In this paper, we present a novel multi-channel ASV model at the frame-level. Specifically, we add spatiotemporal processing blocks (STB) before the pooling layer, which models the contextual relationship within and between channels and across time, respectively. The channel-attended outputs from STB are sent to the pooling layer to obtain an utterance-level speaker representation. Experimental results demonstrate the effectiveness of the proposed method.

show abstract

Section: Methodsmentioning

confidence: 99%

Multi-Channel Far-Field Speaker Verification with Large-Scale Ad-hoc Microphone Arrays

Liang¹,

Chen²,

Yao³

et al. 2021

Preprint

View full text Add to dashboard Cite

show abstract

“…Front-end embedding extractor. The CNN model used as the front-end embedding extractor to train the householdadapted scoring model with VoxCeleb1 data is Half-ResNet34 [23,24], which has half of the channel numbers of the original ResNet34. The model was trained on VoxCeleb2 [8] with 5994 speakers for 100 epochs on a single Tesla V100 Nvidia GPU.…”

Section: Model Trainingmentioning

confidence: 99%

Improving Speaker Identification for Shared Devices by Adapting Embeddings to Speaker Subsets

Tan,

Yang,

Han

et al. 2021

Preprint

View full text Add to dashboard Cite

Speaker identification typically involves three stages. First, a front-end speaker embedding model is trained to embed utterance and speaker profiles. Second, a scoring function is applied between a runtime utterance and each speaker profile. Finally, the speaker is identified using nearest neighbor according to the scoring metric. To better distinguish speakers sharing a device within the same household, we propose a household-adapted nonlinear mapping to a low dimensional space to complement the global scoring metric. The combined scoring function is optimized on labeled or pseudo-labeled speaker utterances. With input dropout, the proposed scoring model reduces EER by 45-71% in simulated households with 2 to 7 hard-to-discriminate speakers per household. On real-world internal data, the EER reduction is 49.2%. From t-SNE visualization, we also show that clusters formed by household-adapted speaker embeddings are more compact and uniformly distributed, compared to clusters formed by global embeddings before adaptation.

show abstract

“…For contrastive learning objectives, following the implementation in [22], we randomly sample M segments from each of N utterances, whose embeddings are xj,i where 1 ≤ j ≤ N and 1 ≤ i ≤ M . The segments sampled from the same utterance are considered from the same speaker and segments from different utterance are considered from different speaker.…”

Section: Contrastive Learning Objectives For Self-supervised Trainingmentioning

confidence: 99%

“…For prototypical loss (ProtoLoss), each mini-batch contains a support set S and a query set Q. Same as the implementation in [22], the M -th segment from each utterance is considered as query. Then the prototype (centroid) is defined as:…”

Section: Angular Prototypical Lossmentioning

confidence: 99%

Self-Supervised Learning Based Domain Adaptation for Robust Speaker Verification

Chen,

Wang,

Qian

2021

Preprint

View full text Add to dashboard Cite

Large performance degradation is often observed for speaker verification systems when applied to a new domain dataset. Given an unlabeled target-domain dataset, unsupervised domain adaptation (UDA) methods, which usually leverage adversarial training strategies, are commonly used to bridge the performance gap caused by the domain mismatch. However, such adversarial training strategy only uses the distribution information of target domain data and can not ensure the performance improvement on the target domain. In this paper, we incorporate self-supervised learning strategy to the unsupervised domain adaptation system and proposed a self-supervised learning based domain adaptation approach (SSDA). Compared to the traditional UDA method, the new SSDA training strategy can fully leverage the potential label information from target domain and adapt the speaker discrimination ability from source domain simultaneously. We evaluated the proposed approach on the Vox-Celeb (labeled source domain) and CnCeleb (unlabeled target domain) datasets, and the best SSDA system obtains 10.2% Equal Error Rate (EER) on the CnCeleb dataset without using any speaker labels on CnCeleb, which also can achieve the state-of-the-art results on this corpus.

show abstract

In defence of metric learning for speaker recognition

Cited by 33 publications

References 39 publications

Multi-Channel Far-Field Speaker Verification with Large-Scale Ad-hoc Microphone Arrays

Multi-Channel Far-Field Speaker Verification with Large-Scale Ad-hoc Microphone Arrays

Improving Speaker Identification for Shared Devices by Adapting Embeddings to Speaker Subsets

Self-Supervised Learning Based Domain Adaptation for Robust Speaker Verification

Contact Info

Product

Resources

About