A new loss function for speaker recognition with deep neural network is proposed, based on Jeffreys Divergence. Adding this divergence to the cross-entropy loss function allows to maximize the target value of the output distribution while smoothing the non-target values. This objective function provides highly discriminative features. Beyond this effect, we propose a theoretical justification of its effectiveness and try to understand how this loss function affects the model, in particular the impact on dataset types (i.e. in-domain or out-of-domain w.r.t the training corpus). Our experiments show that Jeffreys loss consistently outperforms the state-of-the-art for speaker recognition, especially on out-of-domain data, and helps limit false alarms.

International audience

International audience

Attribute-driven privacy aims to conceal a single user’s attribute, contrary to anonymisation that tries to hide the full identity of the user in some data. When the attribute to protect from malicious inferences is binary, perfect privacy requires the log-likelihood-ratio to be zero resulting in no strength-of-evidence. This work presents an approach based on normalizing flow that maps a feature vector into a latent space where the evidence, related to the binary attribute, and an independent residual are disentangled. It can be seen as a non-linear discriminant analysis where the mapping is invertible al-lowing generation by mapping the latent variable back to the original space. This framework allows to manipulate the log-likelihood-ratio of the data and therefore allows to set it to zero for privacy. We show the applicability of the approach on an attribute-driven privacy task where the sex information is removed from speaker embeddings. Results on VoxCeleb2 dataset show the efficiency of the method that outperforms in terms of privacy and utility our previous experiments based on adversarial disentanglement.

International audience

In speaker verification, the extraction of voice representations is mainly based on the Residual Neural Network (ResNet) architecture. ResNet is built upon convolution layers which learn filters to capture local spatial patterns along all the input, then generate feature maps that jointly encode the spatial and channel information. Unfortunately, all feature maps in a convolution layer are learnt independently (the convolution layer does not exploit the dependencies between feature maps) and locally. This problem has first been tackled in image processing. A channel attention mechanism, called squeeze-and-excitation (SE), has recently been proposed in convolution layers and applied to speaker verification. This mechanism re-weights the information extracted across features maps. In this paper, we first propose an original qualitative study about the influence and the role of the SE mechanism applied to the speaker verification task at different stages of the ResNet, and then evaluate several SE architectures. We finally propose to improve the SE approach with a new pooling variant based on the concatenation of mean- and standard-deviation-pooling. Results showed that applying SE only on the first stages of the ResNet allows to better capture speaker information for the verification task, and that significant discrimination gains on Voxceleb1-E, Voxceleb1-H and SITW evaluation tasks have been noted using the proposed pooling variant.

International audience

The I4U consortium was established to facilitate a joint entry to NIST speaker recognition evaluations (SRE). The latest edition of such joint submission was in SRE 2018, in which the I4U submission was among the best-performing systems. SRE'18 also marks the 10-year anniversary of I4U consortium into NIST SRE series of evaluation. The primary objective of the current paper is to summarize the results and lessons learned based on the twelve sub-systems and their fusion submitted to SRE'18. It is also our intention to present a shared view on the advancements, progresses, and major paradigm shifts that we have witnessed as an SRE participant in the past decade from SRE'08 to SRE'18. In this regard, we have seen, among others, a paradigm shift from supervector representation to deep speaker embedding, and a switch of research challenge from channel compensation to domain adaptation.

International audience

18th Annual Conference of the International Speech Communication Association, INTERSPEECH 2017, Stockholm, Sweden, 20-24 August 2017

International audience

Machine Learning (ML) techniques have allowed a great performance improvement of different challenging Spoken Language Understanding (SLU) tasks. Among these methods, Neural Networks (NN), or Multilayer Perceptron (MLP), recently received a great interest from researchers due to their representation capability of complex internal structures in a low dimensional subspace. However, MLPs employ document representations based on basic word level or topic-based features. Therefore, these basic representations reveal little in way of document statistical structure by only considering words or topics contained in the document as a “bag-of-words”, ignoring relations between them. We propose to remedy this weakness by extending the complex features based on Quaternion algebra presented in [1] to neural networks called QMLP. This original QMLP approach is based on hyper-complex algebra to take into consideration features dependencies in documents. New document features, based on the document structure itself, used as input of the QMLP, are also investigated in this paper, in comparison to those initially proposed in [1]. Experiments made on a SLU task from a real framework of human spoken dialogues showed that our QMLP approach associated with the proposed document features outperforms other approaches, with an accuracy gain of 2% with respect to the MLP based on real numbers and more than 3% with respect to the first Quaternion-based features proposed in [1]. We finally demonstrated that less iterations are needed by our QMLP architecture to be efficient and to reach promising accuracies.

State-of-the-art speaker recognition systems performance degrades considerably in noisy environments even though they achieve very good results in clean conditions. In order to deal with this strong limitation, we aim in this work to remove the noisy part of an i-vector directly in the i-vector space. Our approach offers the advantage to operate only at the i-vector extraction level, letting the other steps of the system unchanged. A maximum a posteriori (MAP) procedure is applied in order to obtain clean version of the noisy i-vectors taking advantage of prior knowledge about clean i-vectors distribution. To perform this MAP estimation, Gaussian assumptions over clean and noise i-vectors distributions are made. Operating on NIST 2008 data, we show a relative improvement up to 60% compared with baseline system. Our approach also outperforms the “multi-style” backend training technique. The efficiency of the proposed method is obtained at the price of relative high computational cost. We present at the end some ideas to improve this aspect.

In this paper, we study the impact of dialogue representations and classification methods in the task of theme identification of telephone conversation services having highly imperfect automatic transcriptions. Two dialogue representations are firstly compared: the classical Term Frequency-Inverse Document Frequency with Gini purity criteria (TF-IDF-Gini) method and the Latent Dirichlet Allocation (LDA) approach. We then propose to study an original classification method that takes advantage of the LDA topic space representation, highlighted as the best dialogue representation. To do so, two assumptions about topic representation led us to choose a Gaussian process (GP) based method. This approach is compared with a Support Vector Machine (SVM) classification method. Results show that the GP approach is a better solution to deal with the multiple theme complexity of a dialogue, no matter the conditions studied (manual or automatic transcriptions). We finally discuss the impact of the topic space reduction on the classification accuracy.

I4U is a joint entry of nine research Institutes and Universities across 4 continents to NIST SRE 2012. It started with a brief discussion during the Odyssey 2012 workshop in Singapore. An online discussion group was soon set up, providing a discussion platform for different issues surrounding NIST SRE’12. Noisy test segments, uneven multi-session training, variable enrollment duration, and the issue of open-set identification were actively discussed leading to various solutions integrated to the I4U submission. The joint submission and several of its 17 sub-systems were among top-performing systems. We summarize the lessons learnt from this large-scale effort.

In the field of speaker recognition, the recently proposed notion of “Speaker Binary Key” provides a representation of each acoustic frame in a discriminant binary space. This approach relies on an unique acoustic model composed by a large set of speaker specific local likelihood peaks (called specificities). The model proposes a spatial coverage where each frame is characterized in terms of neighborhood. The most frequent specificities, picked up to represent the whole utterance, generate a binary key vector. The flexibility of this modeling allows to capture non-parametric behaviors. In this paper, we introduce a concept of “typicality” between binary keys, with a discriminant goal. We describe an algorithm able to extract such typicalities, which involves a singular value decomposition in a binary space. The theoretical aspects of this decomposition as well as its potential in terms of future developments are presented. All the propositions are also experimentally validated using NIST SRE 2008 framework.

Short speech duration remains a critical factor of performance degradation when deploying a speaker verification system. To overcome this difficulty, a large number of commercial applications impose the use of fixed pass-phrases. In this context, we show that the performance of the popular i-vector approach can be greatly improved by taking advantage of the phonetic information that they convey. Moreover, as i-vectors require a conditioning process to reach high accuracy, we show that further improvements are possible by taking advantage of this phonetic information within the normalisation process. We compare two methods, Within Class Covariance Normalization (WCCN) and Eigen Factor Radial (EFR), both relying on parameters estimated on the same development data. Our study suggests that WCCN is more robust to data mismatch but less efficient than EFR when the development data has a better match with the test data.

International audience

In supervector UBM/GMM paradigm, each acoustic file is represented by the mean parameters of a GMM model. This supervector space is used as a data representation space, which has a high dimensionality. Moreover, this space is not intrinsically discriminant and a complete speech segment is represented by only one vector, withdrawing mainly the possibility to take into account temporal or sequential information. This work proposes a new approach where each acoustic frame is represented in a discriminant binary space. The proposed approach relies on a UBM to structure the acoustic space in regions. Each region is then populated with a set of Gaussian models, denoted as “specificities”, able to emphasize speaker specific information. Each acoustic frame is mapped in the discriminant binary space, turning “on” or “off” all the specificities to create a large binary vector. All the following steps, speaker reference extraction, likelihood estimation or decision take place in this binary space. Even if this work is a first step in this avenue, the experiments based on NIST SRE 2008 framework demonstrate the potential of the proposed approach. Moreover, this approach opens the opportunity to rethink all the classical processes using a discrete, binary view.

International audience

International audience