Maximum likelihood linear regression for speaker adaptation of continuous density hidden Markov models

By Chris Leggetter, Philip C. Woodland. 1995

How researchers see this paper

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

1. Li Deng
2. Yoshua Bengio
3. Dong Yu
4. Jinyu Li
5. Jianfeng Gao
6. Dacheng Tao
7. Ruslan Salakhutdinov
8. Quoc V. Le

Citations in the corpus

• Agnostic Federated Learning (2019, background): “…in natural language processing (Dredze et al., 2007; Blitzer et al., 2007; Jiang and Zhai, 2007; Raju et al., 2018), speech recognition (Legetter and Woodland, 1995; Gauvain and Chin-Hui, 1994; Pietra et al., 1992; Rosenfeld, 1996; Jelinek, 1998; Roark and Bacchiani, 2003), and…”
• Multiple Source Adaptation and the Rényi Divergence (2009, background): “This problem arises in a variety of natural language processing tasks such parsing, statistical language modeling, text classification [15, 7, 16, 9, 10, 4, 6], or speech processing [11, 8, 14] and computer vision [13] tasks, as well as in many other applications.”
• Maximum a posteriori adaptation of network parameters in deep models (2015, methodology): “The difference is that in MLLR the model parameters are Gaussian mean and variance while here the model parameters are the log-linear model’s transformation matrix weights.”
• Fast speaker adaptation of hybrid NN/HMM model for speech recognition based on discriminative learning of speaker code (2013, background): “In CMLLR this transform can be viewed also as transforming the data itself to match the HMM model.”
• A study of speaker adaptation for DNN-based speech synthesis (2015, methodology): “These techniques can be grouped into two categories: maximum likelihood linear regression (MLLR) [3] and maximum a posteriori (MAP) [4] adaptation.”
• Fast Adaptation of Deep Neural Network Based on Discriminant Codes for Speech Recognition (2014, methodology): “In this work, the fast adaptation scheme is applied to supervised speaker adaptation in speech recognition based on either frame-level cross-entropy or sequence-level maximum mutual information training criterion.”
• Large Vocabulary Continuous Speech Recognition : a ReviewSteve (1996, background): “As more data becomes available, Gaussians can be clustered and an individual transform determined for each cluster[65, 27].”
• Linear Networks Based Speaker Adaptation for Speech Synthesis (2018, methodology): “The Maximum Likelihood Linear Regression (MLLR) is originally developed in automatic speech recognition tasks [2], and it was extended to speech synthesis in [3,4].”
• The LIMSI Broadcast News transcription system (2002, methodology): “Unsupervised acoustic model adaptation (both means and variances) is performed for each cluster using the MLLR technique [ 32 ] after each decoding pass.”
• Learning hidden unit contributions for unsupervised speaker adaptation of neural network acoustic models (2014, background): “The most successful of these is the maximum likelihood linear regression (MLLR) family of techniques developed for GMM-based systems [6, 7, 8].”
• Domain Adaptation in Regression (2011, background): “But, a more challenging problem of domain adaptation arises in a variety of applications, including natural language processing, speech processing, or computer vision [7, 3, 9, 10, 17, 18, 12].”
• Domain Adaptation with Multiple Sources (2008, background): “This set-up and the type of solutions just described have been in fact explored extensively in applications [8, 9, 11, 13–15].”
• Spoofing and countermeasures for speaker verification: A survey (2015, methodology): “maximum likelihood linear regression (MLLR) (Leggetter and Woodland, 1995; Woodland, 2001).”
• Statistical Parametric Speech Synthesis (2007, methodology): “Adaptation (mimicking voices) Techniques of adaptation were originally developed in speech recognition to adjust general acoustic models to a specific speaker or environment to improve the recognition accuracy (Leggetter and Woodland, 1995; Gauvain and Lee, 1994).”
• Developments and directions in speech recognition and understanding, Part 1 [DSP Education] (2009, methodology): “Popular techniques include maximum a posteriori probability (MAP) estimation [17], [38], [51], maximum likelihood linear regression (MLLR) [34], and eigenvoices [30].”