X-Vectors: Robust DNN Embeddings for Speaker Recognition

By David Snyder, Daniel Garcia-Romero, Gregory Sell, Daniel Povey, Sanjeev Khudanpur. 2018

3,060 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Representation & Self-Supervised Learning, Speech & Audio, Signal Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Representation & Self-Supervised Learning

In Speech & Audio

In Signal Processing

Citations in the corpus

The People's Speech: A Large-Scale Diverse English Speech Recognition Dataset for Commercial Usage (2021, background): “We think locality-sensitive hashing [21] of ivectors [24], xvectors [30], or other text-independent speaker embedding is a promising approach to find speaker overlap in utterances.”
VOXLINGUA107: A Dataset for Spoken Language Recognition (2020, methodology): “In all experiments, we use language embedding models derived from the x-vector paradigm [12, 13], with several enhancements.”
CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for Unsegmented Recordings (2020, positive): “It is an x-vector system [47] with a 5-layer TDNN trained on the VoxCeleb dataset [43].”
Ego4D: Around the World in 3,000 Hours of Egocentric Video (2021, background): “[206] David Snyder, Daniel Garcia-Romero, Gregory Sell, Daniel Povey, and Sanjeev Khudanpur.”
Privacy-Preserving Adversarial Representation Learning in ASR: Reality or Illusion? (2019, positive): “We implement the attacker with the state-of-the-art x-vector speaker recognition technique [24].”
UniCATS: A Unified Context-Aware Text-to-Speech Framework with Contextual VQ-Diffusion and Vocoding (2023, methodology): “Therefore, CTX-vec2wav proposes a novel approach of leveraging the mel-spectrogram m to prompt the acoustic contexts, as opposed to conventional methods such as x-vectors (Snyder et al. 2018) or acoustic tokens from audio codec models.”
A3T: Alignment-Aware Acoustic and Text Pretraining for Speech Synthesis and Editing (2022, neutral): “In this setting, we ﬁnd the FastSpeech 2 fails to generate high-quality audio for these new speakers, even equipped with X-Vector (Snyder et al., 2018) to generate speaker embeddings for new speakers.”
SpeechT5: Unified-Modal Encoder-Decoder Pre-Training for Spoken Language Processing (2021, methodology): “To support multi-speaker TTS and VC, the speaker embedding extracted with the x-vector (Snyder et al., 2018) is concatenated with the output of the speech-decoder pre-net followed by a linear layer.”
VoxCeleb2: Deep Speaker Recognition (2018, background): “The success of deep learning in computer vision and speech recognition has motivated the use of deep neural networks (DNN) as feature extractors combined with classiﬁers, though not trained end-to-end [21, 22, 23, 24, 25].”
Icassp 2022 Deep Noise Suppression Challenge (2022, background): “Participants can leverage state-of-the-art speaker models [17, 18, 16] and may choose a speaker embedding extractor which is publicly available [17, 18].”
SUPERB: Speech processing Universal PERformance Benchmark (2021, methodology): “As for ASV , we adopt the well-known x-vector [41] as the downstream model and change Softmax loss to AMSoft-max loss with the same hyper-parameters as [26].”
Zero-Shot Multi-Speaker Text-To-Speech with State-Of-The-Art Neural Speaker Embeddings (2019, methodology): “The clean speech was augmented with reverberation, noise, music, and babble, as described in [17], and then a random subset of these 1,000,000 augmented utterances was combined with the original clean speech.”
Pyannote.Audio: Neural Building Blocks for Speaker Diarization (2019, background): “Speaker embedding uses a slightly wider (512 units instead of 128) and deeper (3 recurrent layers instead of 2) network, relies on x-vector - like statistical temporal pooling [16] and has been trained on shorter 500ms audio chunks.”
SC-GlowTTS: an Efficient Zero-Shot Multi-Speaker Text-To-Speech Model (2021, background): “[9] D. Snyder, D. Garcia-Romero, G. Sell, D. Povey, and S. Khudanpur, “X-vectors: Robust dnn embeddings for speaker recognition,” in 2018 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP).”
Self-supervised Sequence-to-sequence ASR using Unpaired Speech and Text (2019, methodology): “Here, we use 83 dimensional log-Mel filterbank features as targets and x-vectors [22] are fed as auxiliary information to provide speaker characteristics of each utterance.”
Utterance-level Aggregation for Speaker Recognition in the Wild (2019, methodology): “[11] introduces an extension of the method in which the standard deviation is used as well as the mean – this method is termed statistical pooling , and used by [12, 21].”
Exploring Machine Speech Chain for Domain Adaptation and Few-Shot Speaker Adaptation (2021, methodology): “As shown in Figure 1, we use the speaker model [28] pretrained on two large speaker verification datasets [29, 30].”
How Far Are We from Robust Voice Conversion: A Survey (2020, methodology): “X-vector [19] uses a time-delayed neural network (TDNN) to learn temporal context information.”
Approaches to Multi-domain Language Recognition (2018, background): “Embeddings networks have been shown to respond surprisingly well to artificial degradation of audio during training [14, 16].”