Deep neural networks for small footprint text-dependent speaker verification

By Ehsan Variani, X. Lei, Erik McDermott, Ignacio Lopez-Moreno, Javier Gonzalez-Dominguez. 2014

959 citations recorded by the source corpus.

Subfields: Artificial Intelligence, 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

1. Jinyu Li
2. Shinji Watanabe
3. Dong Yu
4. Yoshua Bengio
5. Yu Zhang
6. James R. Glass
7. Hung-yi Lee
8. Li Deng

In Speech & Audio

1. Jinyu Li
2. Shinji Watanabe
3. Yu Zhang
4. Hung-yi Lee
5. Wei-Ning Hsu
6. James R. Glass
7. Yonghui Wu
8. Ye Jia

In Signal Processing

1. Jinyu Li
2. Shinji Watanabe
3. Dong Yu
4. Yu Zhang
5. Yoshua Bengio
6. Hung-yi Lee
7. Daniel Povey
8. James R. Glass

Citations in the corpus

• Transfer Learning from Speaker Verification to Multispeaker Text-To-Speech Synthesis (2018, methodology): “The network maps a sequence of log-mel spectrogram frames computed from a speech utterance of arbitrary length, to a fixed-dimensional embedding vector, known as d-vector [20, 9].”
• Meta-TTS: Meta-Learning for Few-Shot Speaker Adaptive Text-to-Speech (2021, methodology): “D-vector [21], [22] is the utterance-level feature extracted from a pre-trained speaker verification model.”
• Generalized End-to-End Loss for Speaker Verification (2017, background): “In recent years, more efforts have been focusing on using neural networks for speaker verification, while the most successful systems use end-to-end training [8, 9, 10, 11, 12].”
• End-to-end text-dependent speaker verification (2015, background): “Note that compared to [15], we have improved our d-vectors significantly [4].”
• Voice Separation with an Unknown Number of Multiple Speakers (2020, background): “(2018a) proposed to use d-vectors (Variani et al., 2014) as speaker embeddings and directly output the target separated sources.”
• Interpretable Convolutional Filters with SincNet (2018, background): “employed in many state-of-the-art systems [19, 20, 21], directly feeding a CNN with spectrogram bins [22, 23, 24] or even with raw audio samples [25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37] is an approach of increasing popularity.”
• Disentangled Speaker and Language Representations Using Mutual Information Minimization and Domain Adaptation for Cross-Lingual TTS (2021, positive): “Such representations are usually called speaker embedding and language embedding, which can be extracted by a deep neural network (DNN) from acoustic features [7] or merely using a look-up embedding table [8].”
• Teacher-Student Training for Text-Independent Speaker Recognition (2018, methodology): “proposed d-vectors [8] which build speaker embeddings by accumulating frame-level DNN activations across an utterance.”
• Speaker adaptive training in deep neural networks using speaker dependent bottleneck features (2016, methodology): “Speaker normalisation is performed on the input features by transforming the FBANK features with CMLLR transforms (CMLLR-FBANK) and appending speaker information in the form of D-vectors [4].”
• Locally-connected and convolutional neural networks for small footprint speaker recognition (2015, result): “Previous work [4] introduced our baseline system and compared it to the more standard i-vector approach.”
• Wespeaker: A Research and Production Oriented Speaker Embedding Learning Toolkit (2022, background): “Deep speaker embeddings [1, 2, 3, 4] are the de facto standard speaker identity representation for many related tasks.”
• Deep Neural Network Approaches to Speaker and Language Recognition (2015, background): “HE impressive gains in performance obtained using deep neural networks (DNNs) for automatic speech recognition (ASR) [1] have motivated the application of DNNs to other speech technologies such as speaker recognition (SR) and language recognition (LR) [2]–[11].”