For co-channel FM reception with two simultaneously active sources, two-pass constant modulus algorithm (CMA) can provide a coarse decomposition of the overlapped signals, but the weak branch often remains severely distorted after demodulation. We propose TACE-Net, a two-stage asymmetric conditional enhancement framework for weak-source recovery. Stage I refines the dominant CMA branch, and Stage II enhances the weak branch using the pre-CMA mixture, the weak branch, and the dominant branch refined in Stage I. To benchmark weakbranch recovery, we construct VCTK-Radio, a dataset simulating FM modulation, co-channel mixing, CMA-based separation, and demodulation using the VCTK corpus. On VCTK-Radio, TACENet improves DNSMOS-OVRL from 1.164 to 2.901 and PESQ from 1.169 to 1.901, while reducing WER from 71.25% to 28.72% on the weak branch, outperforming competitive baselines.

Conventional audio steganography methods typically require embedding secret information into the carrier, making them vulnerable to steganalysis. To address this issue, we propose a novel coverless audio steganography method that hides information by generating carriers and establishing mapping rules rather than embedding data directly. Our approach leverages a differential privacy clustering algorithm to cluster audio data and select representative audio files, thereby enhancing the security of the steganography. Additionally, we introduce an improved audio feature extraction method that combines traditional Mel-frequency cepstral coefficients (MFCC) with global statistical information, significantly boosting the robustness of the secret information against common audio attacks, particularly time-stretching attacks. Experimental results show that our method achieves a robustness rate of up to 95% against time-stretching and maintains an average security accuracy rate exceeding 97% across various attack scenarios. The proposed method ensures that the audio carrier remains unaltered, thus effectively resisting detection by steganalysis tools. This innovative approach provides a practical and efficient solution for the secure transmission of information in the digital era.

Target speaker extraction aims to isolate a specific speaker’s voice from a composite of multiple sound sources, guided by an enrollment utterance or called anchor. Current methods predominantly derive speaker embeddings from the anchor and integrate them into the separation network to separate the voice of the target speaker. However, the representation of the speaker embedding is too simplistic, often being merely a 1×1024 vector. This dense information makes it difficult for the separation network to harness effectively. To address this limitation, we introduce a pioneering methodology called Hierarchical Representation (HR) that seamlessly fuses anchor data across granular and overarching 5 layers of the separation network, enhancing the precision of target extraction. HR amplifies the efficacy of anchors to improve target speaker isolation. On the Libri-2talker dataset, HR substantially outperforms state-of-the-art time-frequency domain techniques. Further demonstrating HR’s capabilities, we achieved first place in the prestigious ICASSP 2023 Deep Noise Suppression Challenge. The proposed HR methodology shows great promise for advancing target speaker extraction through enhanced anchor utilization.

Subband-based approaches process subbands in parallel through the model with shared parameters to learn the commonality of local spectrums for noise reduction. In this way, they have achieved remarkable results with fewer parameters. However, in some complex environments, the lack of global spectral information has a negative impact on the performance of these subband-based approaches. To this end, this paper introduces the subband interaction as a new way to complement the subband model with the global spectral information such as cross-band dependencies and global spectral patterns, and proposes a new lightweight single-channel speech enhancement framework called Interactive Subband Network (Inter-SubNet). Experimental results on DNS Challenge - Interspeech 2021 dataset show that the proposed Inter-SubNet yields a significant improvement over the subband model and outperforms other state-of-the-art speech enhancement approaches, which demonstrate the effectiveness of subband interaction.

This paper describes the legends-tencent team’s real-time General Speech Restoration (Gesper) system submitted to the ICASSP 2023 Speech Signal Improvement (SSI) Challenge. This newly proposed system is a two-stage architecture, in which the speech restoration is performed, and then followed by speech enhancement. We propose a complex spectral mapping-based generative adversarial network (CSM-GAN) as the speech restoration module for the first time. For noise suppression and dereverberation, the enhancement module is presented with fullband-wideband parallel processing. On the blind test set of ICASSP 2023 SSI Challenge, the proposed Gesper system, which satisfies the real-time condition, achieves 3.27 P.804 overall mean opinion score (MOS) and 3.35 P.835 overall MOS, ranked 1st in both track 1 and track 2.

Most neural network speech enhancement models ignore speech production mathematical models by directly mapping Fourier transform spectrums or waveforms. In this work, we propose a neural source filter network for speech enhancement. Specifically, we use homomorphic signal processing and cepstral analysis to obtain noisy speech’s excitation and vocal tract. Unlike traditional signal processing, we use an attentive recurrent network (ARN) model predicted ratio mask to replace the liftering separation function. Then two convolutional attentive recurrent network (CARN) networks are used to predict the excitation and vocal tract of clean speech, respectively. The system’s output is synthesized from the estimated excitation and vocal. Experiments prove that our proposed method performs better, with SI-SNR improving by 1.363dB compared to FullSubNet.

This paper introduces the Unbeatable Team’s submission to the ICASSP 2023 Deep Noise Suppression (DNS) Challenge. We expand our previous work, TEA-PSE, to its upgraded version – TEA-PSE 3.0. Specifically, TEA-PSE 3.0 incorporates a residual LSTM after squeezed temporal convolution network (S-TCN) to enhance sequence modeling capabilities. Additionally, the local-global representation (LGR) structure is introduced to boost speaker information extraction, and multi-STFT resolution loss is used to effectively capture the time-frequency characteristics of the speech signals. Moreover, retraining methods are employed based on the freeze training strategy to fine-tune the system. According to the official results, TEA-PSE 3.0 ranks 1st in both ICASSP 2023 DNS-Challenge track 1 and track 2.

The scarcity of labeled far-field speech is a constraint for training superior far-field speaker verification systems. In general, fine-tuning the model pre-trained on large-scale near- field speech through a small amount of far-field speech substantially outperforms training from scratch. However, the vanilla fine-tuning suffers from two limitations – catastrophic forgetting and overfitting. In this paper, we propose a weight transfer regularization (WTR) loss to constrain the distance of the weights between the pre-trained model and the fine-tuned model. With the WTR loss, the fine-tuning process takes advantage of the previously acquired discriminative ability from the large-scale near-field speech and avoids catastrophic for- getting. Meanwhile, the analysis based on the PAC-Bayes generalization theory indicates that the WTR loss makes the fine-tuned model have a tighter generalization bound, thus mitigating the overfitting problem. Moreover, three different norm distances for weight transfer are explored, which are L1-norm distance, L2-norm distance, and Max-norm distance. We evaluate the effectiveness of the WTR loss on VoxCeleb (pre-trained) and FFSVC (fine-tuned) datasets. Experimental results show that the distance-based weight transfer fine-tuning strategy significantly outperforms vanilla fine- tuning and other competitive domain adaptation methods.

This paper describes Tencent Ethereal Audio Lab – Northwestern Polytechnical University personalized speech enhancement (TEA-PSE) system submitted to track 2 of the ICASSP 2022 Deep Noise Suppression (DNS) challenge. Our system specifically combines the dual-stage network which is a superior real-time speech enhancement framework with the ECAPA-TDNN speaker embedding network which achieves state-of-the-art performance in speaker verification. The dual-stage network aims to decouple the primal speech enhancement problem into multiple easier sub-problems. Specifically, in stage 1, only the magnitude of the target speech is estimated, which is incorporated with the noisy phase to obtain a coarse complex spectrum estimation. To facilitate the formal estimation, in stage 2, an auxiliary network serves as a post-processing module, where residual noise and interfering speech are further suppressed and the phase information is effectively modified. With the asymmetric loss function to penalize over-suppression, more target speech is preserved, which is helpful for both speech recognition performance and subjective sense of hearing. Our system reaches 3.97 in overall audio quality (OVRL) MOS and 0.69 in word accuracy (WAcc) on the blind test set of the challenge, which outperforms the DNS baseline by 0.57 OVRL and ranks 1st in track 2.

Personalized speech enhancement (PSE) utilizes additional cues like speaker embeddings to remove background noise and interfering speech and extract the speech from target speaker. Previous work, the Tencent-Ethereal-Audio-Lab personalized speech enhancement (TEA-PSE) system, ranked 1st in the ICASSP 2022 deep noise suppression (DNS2022) challenge. In this paper, we expand TEA-PSE to its sub-band version - TEA-PSE 2.0, to reduce computational complexity as well as further improve performance. Specifically, we adopt finite impulse response filter banks and spectrum splitting to reduce computational complexity. We introduce a time frequency convolution module (TFCM) to the system for increasing the receptive field with small convolution kernels. Besides, we explore several training strategies to optimize the two-stage network and investigate various loss functions in the PSE task. TEA-PSE 2.0 significantly outperforms TEA-PSE in both speech enhancement performance and computation complexity. Experimental results on the DNS2022 blind test set show that TEA-PSE 2.0 brings 0.102 OVRL personalized DNSMOS improvement with only 21.9% multiply-accumulate operations compared with the previous TEA-PSE.

Recently, multi-channel speech enhancement has drawn much interest due to the use of spatial information to distinguish target speech from interfering signal. To make full use of spatial information and neural network based masking estimation, we propose a multi-channel denoising neural network - Spatial DCCRN. Firstly, we extend S-DCCRN to multi -channel scenario, aiming at performing cascaded sub-channel and full-channel processing strategy, which can model different channels separately. Moreover, instead of only adopting multi-channel spectrum or concatenating first-channel's magnitude and IPD as the model's inputs, we apply an angle feature extraction module (AFE) to extract frame-level angle feature embeddings, which can help the model to apparently perceive spatial information. Finally, since the phenomenon of residual noise will be more serious when the noise and speech exist in the same time frequency (TF) bin, we particularly design a masking and mapping filtering method to substitute the traditional filter-and-sum operation, with the purpose of cascading coarsely denoising, dereverberation and residual noise suppression. The proposed model, Spatial-DCCRN, has surpassed EaBNet, FasNet as well as several competitive models on the L3DAS22 Challenge dataset. Not only the 3D scenario, Spatial-DCCRN outperforms state-of-the-art (SOTA) model MIMO-UNet by a large margin in multiple evaluation metrics on the multi-channel ConferencingSpeech2021 Challenge dataset. Ablation studies also demonstrate the effectiveness of different contributions.

The ConferencingSpeech 2021 challenge is proposed to stimulate research on far-field multi-channel speech enhancement for video conferencing. The challenge consists of two separate tasks: 1) Task 1 is multi-channel speech enhancement with single microphone array and focusing on practical application with real-time requirement and 2) Task 2 is multi-channel speech enhancement with multiple distributed micro-phone arrays, which is a non-real-time track and does not have any constraints so that participants could explore any algorithms to obtain high speech quality. Targeting the real video conferencing room application, the challenge database was recorded from real speakers and all recording facilities were located by following the real setup of conferencing room. In this challenge, we open-sourced the list of open source clean speech and noise datasets, simulation scripts, and a baseline system for participants to develop their own system. The final ranking of the challenge will be decided by the subjective evaluation which is performed using Absolute Category Ratings (ACR) to estimate Mean Opinion Score (MOS), speech MOS (S-MOS), and noise MOS (N-MOS). This paper describes the challenge, tasks, datasets, subjective evaluation, and challenge results. The baseline system which is a complex ratio mask based neural network and its experimental results are also presented.

Target speech extraction has attracted widespread attention. When microphone arrays are available, the additional spatial information can be helpful in extracting the target speech. We have recently proposed a channel decorrelation (CD) mechanism to extract the inter-channel differential information to enhance the reference channel encoder representation. Although the proposed mechanism has shown promising results for extracting the target speech from mixtures, the extraction performance is still limited by the nature of the original decorrelation theory. In this paper, we propose two methods to broaden the horizon of the original channel decorrelation, by replacing the original softmax-based inter-channel similarity between encoder representations, using an unrolled probability and a normalized cosine-based similarity at the dimensional-level. Moreover, new combination strategies of the CD-based spatial information and target speaker adaptation of parallel encoder outputs are also investigated. Experiments on the reverberant WSJ0 2-mix show that the improved CD can result in more discriminative differential information and the new adaptation strategy is also very effective to improve the target speech extraction.

Speaker verification has been studied mostly under the single-talker condition. It is adversely affected in the presence of interference speakers. Inspired by the study on target speaker extraction, e.g., SpEx, we propose a unified speaker verification framework for both single- and multi-talker speech, that is able to pay selective auditory attention to the target speaker. This target speaker verification (tSV) framework jointly optimizes a speaker attention module and a speaker representation module via multi-task learning. We study four different target speaker embedding schemes under the tSV framework. The experimental results show that all four target speaker embedding schemes significantly outperform other competitive solutions for multi-talker speech. Notably, the best tSV speaker embedding scheme achieves 76.0% and 55.3% relative improvements over the baseline system on the WSJ0-2mix-extr and Libri2Mix corpora in terms of equal-error-rate for 2-talker speech, while the performance of tSV for single-talker speech is on par with that of traditional speaker verification system, that is trained and evaluated under the same single-talker condition.

The INTERSPEECH 2020 Far-Field Speaker Verification Challenge (FFSVC 2020) addresses three different research problems under well-defined conditions: far-field text-dependent speaker verification from single microphone array, far-field textindependent speaker verification from single microphone array, and far-field text-dependent speaker verification from distributed microphone arrays.All three tasks pose a cross-channel challenge to the participants.To simulate the real-life scenario, the enrollment utterances are recorded from close-talk cellphone, while the test utterances are recorded from the far-field microphone arrays.In this paper, we describe the database, the challenge, and the baseline system, which is based on a ResNetbased deep speaker network with cosine similarity scoring.For a given utterance, the speaker embeddings of different channels are equally averaged as the final embedding.The baseline system achieves minDCFs of 0.62, 0.66, and 0.64 and EERs of 6.27%, 6.55%, and 7.18% for task 1, task 2, and task 3, respectively.

Speaker extraction aims to mimic humans' selective auditory attention by extracting a target speaker's voice from a multi-talker environment. It is common to perform the extraction in frequency-domain, and reconstruct the time-domain signal from the extracted magnitude and estimated phase spectra. However, such an approach is adversely affected by the inherent difficulty of phase estimation. Inspired by Conv-TasNet, we propose a time-domain speaker extraction network (SpEx) that converts the mixture speech into multi-scale embedding coefficients instead of decomposing the speech signal into magnitude and phase spectra. In this way, we avoid phase estimation. The SpEx network consists of four network components, namely speaker encoder, speech encoder, speaker extractor, and speech decoder. Specifically, the speech encoder converts the mixture speech into multi-scale embedding coefficients, the speaker encoder learns to represent the target speaker with a speaker embedding. The speaker extractor takes the multi-scale embedding coefficients and target speaker embedding as input and estimates a receptive mask. Finally, the speech decoder reconstructs the target speaker's speech from the masked embedding coefficients. We also propose a multi-task learning framework and a multi-scale embedding implementation. Experimental results show that the proposed SpEx achieves 37.3%, 37.7% and 15.0% relative improvements over the best baseline in terms of signal-to-distortion ratio (SDR), scale-invariant SDR (SI-SDR), and perceptual evaluation of speech quality (PESQ) under an open evaluation condition.

Speaker extraction is to extract a target speaker's voice from multi-talker speech. It simulates humans' cocktail party effect or the selective listening ability. The prior work mostly performs speaker extraction in frequency domain, then reconstructs the signal with some phase approximation. The inaccuracy of phase estimation is inherent to the frequency domain processing, that affects the quality of signal reconstruction. In this paper, we propose a time-domain speaker extraction network (TseNet) that doesn't decompose the speech signal into magnitude and phase spectrums, therefore, doesn't require phase estimation. The TseNet consists of a stack of dilated depthwise separable convolutional networks, that capture the long-range dependency of the speech signal with a manageable number of parameters. It is also conditioned on a reference voice from the target speaker, that is characterized by speaker i-vector, to perform the selective listening to the target speaker. Experiments show that the proposed TseNet achieves 16.3% and 7.0% relative improvements over the baseline in terms of signal-to-distortion ratio (SDR) and perceptual evaluation of speech quality (PESQ) under open evaluation condition.

The SpeakerBeam-FE (SBF) method is proposed for speaker extraction. It attempts to overcome the problem of unknown number of speakers in an audio recording during source separation. The mask approximation loss of SBF is sub-optimal, which doesn't calculate direct signal reconstruction error and consider the speech context. To address these problems, this paper proposes a magnitude and temporal spectrum approximation loss to estimate a phase sensitive mask for the target speaker with the speaker characteristics. Moreover, this paper explores a concatenation framework instead of the context adaptive deep neural network in the SBF method to encode a speaker embedding into the mask estimation network. Experimental results under open evaluation condition show that the proposed method achieves 70.4% and 17.7% relative improvement over the SBF baseline on signal-to-distortion ratio (SDR) and perceptual evaluation of speech quality (PESQ), respectively. A further analysis demonstrates 69.1% and 72.3% relative SDR improvements obtained by the proposed method for different and same gender mixtures.

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.

The i-vector approach to speaker recognition has achieved good performance when the domain of the evaluation dataset is similar to that of the training dataset. However, in realworld applications, there is always a mismatch between the training and evaluation datasets, that leads to performance degradation. To address this problem, this paper proposes to learn the domain-invariant and speaker-discriminative speech representations via domain adversarial training. Specifically, with domain adversarial training method, we use a gradient reversal layer to remove the domain variation and project the different domain data into the same subspace. Moreover, we compare the proposed method with other state-of-the-art unsupervised domain adaptation techniques for i-vector approach to speaker recognition (e.g. autoencoder based domain adaptation, inter dataset variability compensation, dataset-invariant covariance normalization, and so on). Experiments on 2013 domain adaptation challenge (DAC) dataset demonstrate that the proposed method is not only effective in solving the dataset mismatch problem, but also outperforms the compared unsupervised domain adaptation methods.

Utterance level permutation invariant training (uPIT) technique is a state-of-the-art deep learning architecture for speaker independent multi-talker separation. uPIT solves the label ambiguity problem by minimizing the mean square error (MSE) over all permutations between outputs and targets. However, uPIT may be sub-optimal at segmental level because the optimization is not calculated over the individual frames. In this paper, we propose a constrained uPIT (cuPIT) to solve this problem by computing a weighted MSE loss using dynamic information (i.e., delta and acceleration). The weighted loss ensures the temporal continuity of output frames with the same speaker. Inspired by the heuristics (i.e., vocal tract continuity) in computational auditory scene analysis, we then extend the model by adding a Grid LSTM layer, that we name it as cuPIT-Grid LSTM, to automatically learn both temporal and spectral patterns over the input magnitude spectrum simultaneously. The experimental results show 9.6% and 8.5% relative improvements on WSJ0-2mix dataset under both closed and open conditions comparing with the uPIT baseline.

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

Technical report for NIST LRE 2015 Workshop

Gaussian PLDA with uncertainty propagation is effective for i-vector based speaker verification. The idea is to propagate the uncertainty of i-vectors caused by the duration variability of utterances to the PLDA model. However, a limitation of the method is the difficulty of performing length normalization on the posterior covariance matrix of an i-vector. This paper proposes a method to avoid performing length normalization on i-vectors in Gaussian PLDA modeling so that uncertainty propagation can be directly applied without transforming the posterior covariance matrices of i-vectors. Instead of performing length normalization on i-vectors independently, the proposed method normalizes the column vectors of the total variability matrix. Because the i-vectors of all utterances are derived from the same normalized total variability matrix, they will be subject to the same degree of normalization, thereby avoiding the undesirable distortion introduced by the utterance-dependent length-normalization process. Experimental results on both NIST 2010 and 2012 SREs demonstrate that the proposed method achieves a performance similar to (and in some situations better than) that of Gaussian PLDA with length normalization. The method has the potential of improving the performance of uncertainty propagation for i-vector/PLDA speaker verification.

Conventional PLDA scoring in i-vector speaker verification involves the i-vectors of target speakers and claimants only. We have previously demonstrated that better performance can be achieved by incorporating the information of background speakers in the scoring process via speaker-dependent SVMs. This is achieved by defining a PLDA score space with dimension equal to the number of training i-vectors for each target speaker. The new protocol in NIST 2012 SRE permits systems to use the information of other target-speakers (called known non-targets) in each verification trial. In this paper, we exploit this new protocol to enhance the performance of PLDA-SVM scoring by using the score vectors of both known and unknown non-targets as the impostor class data to train the speaker-dependent SVMs. Because some target speakers have one enrollment utterance only, which results in severe imbalance in the speaker- and impostor-class data for SVM training. This paper shows that if the enrollment utterance is sufficiently long, a number of target-speaker i-vectors can be generated by an utterance partitioning and resampling technique, resulting in much better scoring SVMs. Results on NIST 2012 SRE demonstrate the advantages of pooling the known and unknown non-targets for training the SVMs and that the resampling techniques can help the SVM training algorithm to find better decision boundaries for those speakers with only a small number of enrollment utterances.

15th Annual Conference of the International Speech Communication Association: Celebrating the Diversity of Spoken Languages, INTERSPEECH 2014, 14-18 September 2014

Likelihood ratio (LR) scoring in PLDA speaker verification systems only uses the information of background speakers implicitly. This paper exploits the notion of empirical kernel maps to incorporate background speaker information into the scoring process explicitly. This is achieved by training a scoring SVM for each target speaker based on a kernel in the empirical feature space. More specially, given a test i-vector and the identity of the target under test, a score vector is constructed by computing the LR scores of the test i-vector with respect to the target-speaker's i-vectors and a set of background-speakers' i-vectors. While in most situations, only one target-speaker i-vector is available for training the SVM, this paper demonstrates that if the enrollment utterance is sufficiently long, a number of target-speaker i-vectors can be generated by an utterance partitioning and resampling technique, resulting in much better scoring SVMs. Results on NIST 2010 SRE suggests that the idea of incorporating background speaker information into PLDA scoring through training speaker-dependent SVMs together with the utterance partitioning techniques can boost the performance of i-vector based PLDA systems significantly.

The success of the recent i-vector approach to speaker verification relies on the capability of i-vectors to capture speaker characteristics and the subsequent channel compensation methods to suppress channel variability. Typically, given an utterance, an i-vector is determined from the utterance regardless of its length. This paper investigates how the utterance length affects the discriminative power of i-vectors and demonstrates that the discriminative power of i-vectors reaches a plateau quickly when the utterance length increases. This observation suggests that it is possible to make the best use of a long conversation by partitioning it into a number of sub-utterances so that more i-vectors can be produced for each conversation. To increase the number of sub-utterances without scarifying the representation power of the corresponding i-vectors, repeated applications of frame-index randomization and utterance partitioning are performed. Results on NIST 2010 speaker recognition evaluation (SRE) suggest that (1) using more i-vectors per conversation can help to find more robust linear discriminant analysis (LDA) and within-class covariance normalization (WCCN) transformation matrices, especially when the number of conversations per training speaker is limited; and (2) increasing the number of i-vectors per target speaker helps the i-vector based support vector machines (SVM) to find better decision boundaries, thus making SVM scoring outperforms cosine distance scoring by 19% and 9% in terms of minimum normalized DCF and EER.

This paper presents the HKCUPU speaker recognition system submitted to NIST 2010 speaker recognition evaluation (SRE). The system comprises five subsystems, each with different acoustic features, session-variability reduction methods, speaker modeling and scoring methods and classifiers. This paper reports the results of individual and fusion systems for the core test and highlights the improvements made by our newly proposed JFA-Fishervoice (FSH) subsystem. Results show that FSH outperforms JFA when its projection matrix is channel-dependent (telephone or microphone) and that FSH is complementary to other state-of-the-art techniques. It was also found that VAD is an important pre-processing step for interview speech.

GMM-SVM has become a promising approach to textindependent speaker verification. However, a problematic issue of this approach is the extremely serious imbalance between the numbers of speaker-class and impostor-class utterances available for training the speaker-dependent SVMs. This data-imbalance problem can be addressed by (1) creating more speaker-class supervectors for SVM training through utterance partitioning with acoustic vector resampling (UP-AVR) and (2) avoiding the SVM training so that speaker scores are formulated as an inner product discriminant function (IPDF) between the target-speaker’s supervector and test supervector. This paper highlights the differences between these two approaches and compares the effect of using different kernels – including the KL divergence kernel, GMM-UBM mean interval (GUMI) kernel and geometric-mean-comparison kernel – on their performance. Experiments on the NIST 2010 Speaker Recognition Evaluation suggest that GMM-SVM with UP-AVR is superior to speaker comparison and that the GUMI kernel is slightly better than the KL kernel in speaker comparison. Index Terms: speaker verification, GMM-SVM, speaker comparison, NIST SRE, utterance partitioning, data imbalance.

Using GMM-supervectors as the input to SVM classifiers (namely, GMM-SVM) is one of the promising approaches to text-independent speaker verification. However, one unaddressed issue of this approach is the severe imbalance between the numbers of speaker-class utterances and impostor-class utterances available for training a speaker-dependent SVM. This paper proposes a resampling technique – namely utterance partitioning with acoustic vector resampling (UP-AVR) – to mitigate the data imbalance problem. Specifically, the sequence order of acoustic vectors in an enrollment utterance is first randomized; then the randomized sequence is partitioned into a number of segments. Each of these segments is then used to produce a GMM-supervector via MAP adaptation and mean vector concatenation. A desirable number of speaker-class supervectors can be produced by repeating this randomization and partitioning process a number of times. Experimental evaluations suggest that UP-AVR can reduce the EER of GMM-SVM systems by about 10%. 1.