Support vector machines versus fast scoring in the low-dimensional total variability space for speaker verification

By Najim Dehak, Réda Dehak, Patrick Kenny, Niko Brummer, Pierre Ouellet, Pierre Dumouchel. 2009

392 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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 Signal Processing

Citations in the corpus

Voice Separation with an Unknown Number of Multiple Speakers (2020, methodology): “In (Zmolikova et al., 2017; Delcroix et al., 2018) a neural network employed estimated i-vectors (Dehak et al., 2009) in order to estimate masks, which extract the target speaker by generating beamformer filters, while Wang et al. (2018a) proposed to use d-vectors (Variani et al., 2014) as speaker…”
Unsupervised Learning of Disentangled and Interpretable Representations from Sequential Data (2017, methodology): “For all experiments, we use the fast scoring approach in [4] that uses cosine similarity as the similarity metric and compute the equal error rate (EER).”
Disentangled Sequential Autoencoder (2018, background): “The claimed identity is conﬁrmed if the cosine similarity cos ( w test , w target ) is grater than a given threshold (cid:15) (Dehak et al., 2009).”
Audio Word2Vec: Unsupervised Learning of Audio Segment Representations Using Sequence-to-Sequence Autoencoder (2016, methodology): “It is common to use i-vectors to represent utterances in speaker identification [4], and several approaches have been successfully used in STD [10, 6, 7, 8].”
Neural Attention Models for Sequence Classification: Analysis and Application to Key Term Extraction and Dialogue Act Detection (2016, background): “Many common tasks can be formulated as sequence classification including speaker recognition [10], audio emotion classification [11], spoken term detection (STD) [12–14], dialogue act detection [15–17], key term extraction [18–21], etc.”
Completely Unsupervised Speech Recognition By A Generative Adversarial Network Harmonized With Iteratively Refined Hidden Markov Models (2019, background): “Substantial effort was made to learn signal representations directly from speech signals in an unsupervised way, which could be a good step towards unsupervised ASR, and has been shown useful in various tasks, such as speaker identification [3], spoken term detection [4, 5, 6], and spoken document r”
Speaker Adaptive Training of Deep Neural Network Acoustic Models Using I-Vectors (2015, methodology): “[20], [21] have been explored for this purpose.”
Front-End Factor Analysis for Speaker Verification (2011, methodology): “Based on this, we proposed a new speaker verification system based on factor analysis as a feature extractor [9].”
Analysis of i-vector Length Normalization in Speaker Recognition Systems (2011, background): “As noted in [5], the success of cosine scoring [2] suggests that there is a principal axis of channel variation that is dependent on the speaker identity.”
New Developments in Voice Biometrics for User Authentication (2011, methodology): “The first one is fusing a fourth system which is an i-vector [8] based system which has accuracy comparable to the JFA system.”
The YLI-MED Corpus: Characteristics, Procedures, and Plans (2015, methodology): “The system we chose for the comparison is based on an i-vector approach initially developed by Dehak et al. (2009), with an improvement by Burget et al. (2011).”
On speaker adaptation of long short-term memory recurrent neural networks (2015, methodology): “For the future work, we will study the incorporation of speaker i-vectors [31] for adaptation of LSTM-RNNs.”
Bayesian Speaker Verification with Heavy-Tailed Priors (2010, methodology): “A recent advance in this direction was reported in [1, 2], where a type of principal components analysis was used to represent a given speech utterance by a feature vector of low dimension (400), independent of the length of the utterance.”
Completely Unsupervised Phoneme Recognition by Adversarially Learning Mapping Relationships from Audio Embeddings (2018, background): “These vector representations have been shown useful for many applications, such as speaker identification [13] and audio emotion classification [14], in addition to spoken term detection (STD) [15, 16, 17], because such vector representations can be applied to standard classifiers to determine the s”
Language Recognition via i-vectors and Dimensionality Reduction (2011, methodology): “Speech is windowed at 20ms with a 10ms frame rate filtered through a mel-scale filter bank and then RASTA.”
Speaker recognition by means of Deep Belief Networks (2013, methodology): “Our reference models in this work are the state-of-the-art speaker Gaussian Mixture Models (GMMs), which are used to estimate statistics that allow obtaining a low–dimensional representation of a speech segment, the so–called “identity vector” or ivector [13][14].”