Proposes DNN-based speaker embedding correlated with subjective inter-speaker similarity for speech synthesis.
problem Inadequate speaker representation for open speakers not in training data.
method Two training algorithms using inter-speaker similarity matrices: similarity vector embedding and similarity matrix embedding.
result Proposed algorithms learn speaker embedding highly correlated with subjective inter-speaker similarity.
This paper presents a study on discriminative artificial neural network classifiers in the context of open-set speaker identification. Both 2-class and multi-class architectures are tested against the conventional Gaussian mixture model based classifier on enrolled speaker sets of different sizes. The performance evalu…
Study improves speaker verification accuracy using angular based embedding learning.
problem Improving discriminative power of embeddings for open-set speaker verification.
method Optimizes angular distance and adds margin penalty, applying various angular margin embedding strategies and proposing inter-class regularization.
result Achieved impressive results with 16.5% improvement in EER and 18.2% improvement in minimum detection cost function.
Deep learning improves speaker recognition verification and identification.
problem Limited progress in speaker recognition for 5-6 years.
method Applied deep learning techniques in speaker verification and identification.
result Deep learning becomes the state-of-the-art solution for speaker recognition.
This paper examines the speaker identification potential of breath sounds in continuous speech. Speech is largely produced during exhalation. In order to replenish air in the lungs, speakers must periodically inhale. When inhalation occurs in the midst of continuous speech, it is generally through the mouth. Intra-spee…
Study compares metric learning loss functions for speaker verification.
problem Comparing metric learning loss functions for end-to-end speaker verification.
method Cross entropy loss, cosine loss, angular margin loss, center loss, contrastive loss, triplet loss.
result Additive angular margin loss outperforms other loss functions.
NIST's CTS Challenge evaluates speaker recognition using telephony data.
problem Improving speaker recognition accuracy using telephony data.
method Large-scale neural network architectures, extensive data augmentation, proprietary data, and fine-tuning.
result Remarkable improvements in speaker recognition performance.
When the available data of a target speaker is insufficient to train a high quality speaker-dependent neural text-to-speech (TTS) system, we can combine data from multiple speakers and train a multi-speaker TTS model instead. Many studies have shown that neural multi-speaker TTS model trained with a small amount data f…
In this paper, we present a novel system that separates the voice of a target speaker from multi-speaker signals, by making use of a reference signal from the target speaker. We achieve this by training two separate neural networks: (1) A speaker recognition network that produces speaker-discriminative embeddings; (2) …
The paper proposes deep normalization to improve speaker recognition performance.
problem Non-Gaussian and non-homogeneous distributions of deep speaker vectors negatively impact speaker recognition.
method Proposes a deep normalization approach based on a novel discriminative normalization flow (DNF) model.
result DNF-based normalization delivers substantial performance gains and strong generalization capability.
Speaker embedding models that utilize neural networks to map utterances to a space where distances reflect similarity between speakers have driven recent progress in the speaker recognition task. However, there is still a significant performance gap between recognizing speakers in the training set and unseen speakers. …
End-to-end speaker verification framework reduces text dependency.
problem Improving text-independent speaker verification.
method Jointly trains SE and ASR networks with triplet loss and adversarial gradient.
result Lower equal error rate and better text-independency compared to other approaches.
Graph neural networks refine speaker embeddings for better session-level diarization.
problem Local speaker distinction in meeting sessions using deep embeddings.
method Graph Neural Networks (GNNs) refine speaker embeddings using session-level structural information.
result Spectral clustering on refined embeddings outperforms original embeddings significantly.
Improved neural speaker embeddings enhance ASR performance.
problem Few studies have explored neural speaker embeddings for ASR.
method Integrating improved neural speaker embeddings into a conformer-based hybrid HMM ASR system.
result Improved neural embeddings achieve on-par performance with i-vectors.
Paper improves speaker verification with federated learning and differential privacy.
problem Improving speaker verification accuracy using private data.
method Combining federated learning and differential privacy to train an auxiliary model that predicts vocal characteristics.
result 6% relative improvement in equal error rate over a baseline system.
This paper proposes a new architecture for speaker adaptation of multi-speaker neural-network speech synthesis systems, in which an unseen speaker's voice can be built using a relatively small amount of speech data without transcriptions. This is sometimes called "unsupervised speaker adaptation". More specifically, we…
VoxCeleb 2019 challenge assesses speaker recognition in uncontrolled settings.
problem Evaluate speaker recognition technology in unconstrained data.
method Public dataset, challenge, and workshop at Interspeech 2019.
result Baseline results and discussions provided.
Paper explores using EEG for better speaker identification, even in noisy environments.
problem Speaker identification performance degrades in background noise.
method Uses EEG signals to enhance speaker identification systems, comparing with acoustic features.
result Speaker identification system using only EEG features outperforms one using only acoustic features in high background noise.
Personal VAD detects target speaker voice activity efficiently.
problem Efficiently detect target speaker voice activity for reduced computational cost and battery usage.
method Trains a neural network conditioned on speaker embedding or verification score, outputs probabilities for three speech classes.
result Trained model with 130K parameters outperforms combined standard VAD and speaker recognition networks.
A new type of End-to-End system for text-dependent speaker verification is presented in this paper. Previously, using the phonetically discriminative/speaker discriminative DNNs as feature extractors for speaker verification has shown promising results. The extracted frame-level (DNN bottleneck, posterior or d-vector) …
New models extrapolate false alarms in ASV without new data.
problem Reliable extrapolation of false alarm rates in ASV without new speaker data.
method Generative models in ASV score space for arbitrary systems.
result Models accurately extrapolate false alarm rates for large speaker populations.
Improves speaker verification for variable-duration utterances using a feature pyramid module.
problem Improving robustness for variable-duration utterances in speaker verification.
method Integrates a feature pyramid module into multi-scale aggregation to enhance speaker-discriminative information from multiple layers.
result Improves performance for both short and long utterances compared to state-of-the-art approaches.
The social media revolution has produced a plethora of web services to which users can easily upload and share multimedia documents. Despite the popularity and convenience of such services, the sharing of such inherently personal data, including speech data, raises obvious security and privacy concerns. In particular, …
CAT is a new ASR toolkit using CRF and CTC for state-of-the-art speech recognition.
problem Improving automatic speech recognition systems.
method CRF-based discriminative training with CTC-inspired state topology.
result CAT achieves state-of-the-art results with fewer parameters and is competitive with hybrid models.
New method uses Multiple Choice Learning for speech separation.
problem Ambiguous task of assigning model predictions to ground truth signals.
method Uses Multiple Choice Learning (MCL) instead of Permutation Invariant Training (PIT).
result MCL matches PIT performance but is computationally advantageous.
New method separates multiple voices in mixed audio.
problem Separating multiple simultaneous speakers in audio.
method Gated neural networks trained at multiple steps, selecting actual number of speakers.
result Outperforms current state of the art for more than two speakers.
BOFFIN TTS optimizes hyper-parameters for new speaker adaptation.
problem Fine-tuning a pre-trained TTS model for a new speaker with limited data.
method Bayesian optimization to efficiently find optimal hyper-parameters.
result Average 30% improvement in speaker similarity over standard techniques.
Improved multi-user VoiceFilter-Lite model for speech recognition.
problem Limited performance of multi-user VoiceFilter-Lite models.
method Dual learning rate schedule and FiLM for feature conditioning.
result Closed the performance gap between multi-user and single-user VoiceFilter-Lite models.
Improved far-field speaker verification for short utterances in noisy conditions.
problem Challenges in speaker verification on short utterances in uncontrolled noisy environments.
method Used deep neural network architectures (TDNN and ResNet) and experimented with various embedding extractors and training procedures.
result ResNet architectures outperform x-vector approach in speaker verification quality for both long and short utterances.
Bayesian approach improves speech recognition with limited speaker data.
problem Reduces mismatch between training and evaluation data due to speaker differences.
method Bayesian learning for DNN adaptation models with limited speaker data.
result Bayesian adaptation consistently outperforms deterministic methods, reducing word error rates up to 1.4%.
Most neural-network based speaker-adaptive acoustic models for speech synthesis can be categorized into either layer-based or input-code approaches. Although both approaches have their own pros and cons, most existing works on speaker adaptation focus on improving one or the other. In this paper, after we first systema…
We present a meta-learning approach for adaptive text-to-speech (TTS) with few data. During training, we learn a multi-speaker model using a shared conditional WaveNet core and independent learned embeddings for each speaker. The aim of training is not to produce a neural network with fixed weights, which is then deplo…
Study shows emotion affects speaker recognition and vice versa.
problem Dependencies between emotion and speaker recognition.
method Transfer learning and fine-tuning for emotion classification.
result Fine-tuning improves emotion recognition performance by 30.40% on IEMOCAP, 7.99% on MSP-Podcast, and 8.61% on Crema-D.
Recently, voice conversion (VC) without parallel data has been successfully adapted to multi-target scenario in which a single model is trained to convert the input voice to many different speakers. However, such model suffers from the limitation that it can only convert the voice to the speakers in the training data, …
The human auditory system is able to distinguish the vocal source of thousands of speakers, yet not much is known about what features the auditory system uses to do this. Fourier Transforms are capable of capturing the pitch and harmonic structure of the speaker but this alone proves insufficient at identifying speaker…
In this paper, we propose a fully supervised speaker diarization approach, named unbounded interleaved-state recurrent neural networks (UIS-RNN). Given extracted speaker-discriminative embeddings (a.k.a. d-vectors) from input utterances, each individual speaker is modeled by a parameter-sharing RNN, while the RNN state…
Speech enhancement improved by adapting to unknown speakers without auxiliary signals.
problem Improving speech enhancement accuracy for unknown speakers.
method Adopting multi-task learning for speech enhancement and speaker identification, using multi-head self-attention.
result Achieved state-of-the-art performance and improved subjective quality.
Study speaker verification security using hierarchical Bayesian modeling.
problem Estimating false alarm rate in ASV systems for large speaker databases.
method Hierarchical Bayesian modeling of ASV scores to assess security against closest impostors.
result Neither i-vector nor x-vector systems are secure against increased impostor database size.
In this paper, we propose a new pooling method called spatial pyramid encoding (SPE) to generate speaker embeddings for text-independent speaker verification. We first partition the output feature maps from a deep residual network (ResNet) into increasingly fine sub-regions and extract speaker embeddings from each sub-…
WEEND uses a neural network to recognize speech and assign speakers to words.
problem End-to-end neural diarization without additional ASR and orchestration.
method Multi-task learning with an auxiliary network for ASR and speaker diarization.
result WEEND outperforms turn-based diarization and can handle 5-minute audio.
In this paper we study speaker linking (a.k.a.\ partitioning) given constraints of the distribution of speaker identities over speech recordings. Specifically, we show that the intractable partitioning problem becomes tractable when the constraints pre-partition the data in smaller cliques with non-overlapping speakers…
Study combines speaker verification and voice trigger detection in a single network.
problem Separate training for speaker verification and voice trigger detection.
method Multi-task learning with a single network trained on both tasks.
result Single network achieves comparable accuracy to independent models for each task.
For many years, i-vector based audio embedding techniques were the dominant approach for speaker verification and speaker diarization applications. However, mirroring the rise of deep learning in various domains, neural network based audio embeddings, also known as d-vectors, have consistently demonstrated superior spe…
Meta-learning framework improves short utterance speaker recognition.
problem Poor performance of existing models with short utterances.
method Prototypical Networks with support and query sets, enforcing classification against entire training set.
result Significant performance gains on VoxCeleb datasets.
Self multi-head attention improves speaker recognition for long utterances.
problem Speaker recognition for long speech segments using Deep Learning.
method Convolutional Neural Network (CNN) for short-term features, self multi-head attention for long-term embeddings.
result Self multi-head attention outperforms other pooling methods by 18% relative EER on VoxCeleb1 dataset.
We investigate deep neural network performance in the textindependent speaker recognition task. We demonstrate that using angular softmax activation at the last classification layer of a classification neural network instead of a simple softmax activation allows to train a more generalized discriminative speaker embedd…
NIST CTS Superset offers a large dataset for telephony speaker recognition.
problem Lack of a large-scale, uniform dataset for telephony speaker recognition.
method Compilation of speech segments from multiple corpora, including Greybeard, Switchboard, and Mixer series.
result Results on the NIST 2020 CTS Speaker Recognition Challenge serve as a reference baseline.
The NL score optimizes speaker recognition tasks.
problem Improving speaker recognition accuracy.
method Established the theory of optimal scores based on normalized likelihood.
result NL score is equivalent to PLDA likelihood ratio under certain conditions.