Speaker-adaptation for hybrid HMM-ANN continuous speech recognition system

By João Paulo da Silva Neto, L. Almeida, M. Hochberg, Ciro Martins, Luís Nunes, S. Renals, Tony A. Robinson. 1995

It is well known that recognition performance degrades signi cantly when moving from a speakerdependent to a speaker-independent system. Traditional hidden Markov model (HMM) systems have successfully applied speaker-adaptation approaches to reduce this degradation. In this paper we present and evaluate some techniques for speaker-adaptation of a hybrid HMM-arti cial neural network (ANN) continuous speech recognition system. These techniques are applied to a well trained, speaker-independent, hybrid HMM-ANN system and the recognizer parameters are adapted to a new speaker through o -line procedures. The techniques are evaluated on the DARPA RM corpus using varying amounts of adaptation material and different ANN architectures. The results show that speaker-adaptation within the hybrid framework can substantially improve system performance.

Subfields: Speech & Audio

How researchers see this paper

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

In Speech & Audio

Citations in the corpus

Unsupervised Pre-Training of Bidirectional Speech Encoders via Masked Reconstruction (2020, methodology): “For domain adaptation we use linear input network (LIN, [29, 30]), which inserts an additional linear layer (initialized as the identity mapping) between the input and the pre-trained network, and only adapts this layer and the soft-max layer for the ﬁrst 5 epochs of supervised training.”
Context-Dependent Pre-Trained Deep Neural Networks for Large-Vocabulary Speech Recognition (2012, methodology): “Inspiration for such algorithms may come from the ANN-HMM literature (e.g., [72] and [73]) or the many successful adaptation techniques developed in the past decades for GMM-HMMs (e.g., MLLR [74], MAP [75], joint compensation of distortions [76], variable parameter HMMs [77]).”
Multi-basis adaptive neural network for rapid adaptation in speech recognition (2015, methodology): “In addition, the transformation based schemes treat the SI neural network as canonical model and add additional linear hidden layers as speaker-dependent transforms prior to the input layer [16, 17], to the hidden layer [18, 19], or to the output layer [20].”
Maximum a posteriori adaptation of network parameters in deep models (2015, background): “Recent approaches, such as regularization based [7, 8], subspace based [9, 10], transformation based [6, 11, 12, 13], i-Vector based [14], native neural network based [15, 16, 17], factorization based [18] and fast adaptation schemes based on discriminant speaker codes [19, 16, 20, 21], have been pr”
Connectionist speech recognition of Broadcast News (2002, methodology): “Current approaches to connectionist speaker adaptation are based on gradient-descent (Neto et al., 1995), although Fritsch et al. (1998) have recently presented a promising method using a somewhat different connectionist acoustic model.”
Low-rank plus diagonal adaptation for deep neural networks (2016, methodology): “A simple and popular approach to combat the mismatch problem is to adapt the DNN model by applying a linear transformation on either input features, hidden layers, or the input to the softmax layer [5, 6, 7, 8, 9].”
Comparison of discriminative input and output transformations for speaker adaptation in the hybrid NN/HMM systems (2010, methodology): “Techniques, like Linear Input Network (LIN)[2], Retrained Speaker-Independent (RSI), Parallel Hidden Network (PHN) and Gamma Network (GAMMA) for adapting NN are compared in [3].”
Modeling speaker variability using long short-term memory networks for speech recognition (2015, background): “Alternatively, a group of approaches perform adaptation by adding a linear transformation layer to the input [7], hidden [8], or output layer [9] of a DNN.”
Unsupervised Adaptation of Acoustic Models for ASR Using Utterance-Level Embeddings from Squeeze and Excitation Networks (2019, background): “In case of speaker adaptation, speaker-specific linear layers can be inserted either in input, output or in hidden layers [2],[3],[4].”
Feature space maximum a posteriori linear regression for adaptation of deep neural networks (2014, methodology): “In this paper, a feature space maximum a posteriori (MAP) linear regression (LR) framework is proposed to adapt the parameters of CD-DNN-HMMs based on a linear transformation network (LIN) [18, 19].”
A Study of Enhancement, Augmentation, and Autoencoder Methods for Domain Adaptation in Distant Speech Recognition (2018, background): “Note that we do not consider the online adaptive setting, a common scenario for speaker adaptation [3, 4, 20, 21], where we have a small amount of labeled data to adapt to the target domain.”
Machine Learning Paradigms for Speech Recognition: An Overview (2013, methodology): “[194], [195], the target MLP is obtained by augmenting the”
Linear Networks Based Speaker Adaptation for Speech Synthesis (2018, methodology): “According to the different positions of LN, LN based adaptation method includes Linear Input Network (LIN [17]), Linear Hidden Network (LHN [18]) and Linear Output Network (LON [19]), while LHN can be inserted to any positions between two successive hidden layers.”
Adaptation of context-dependent deep neural networks for automatic speech recognition (2012, methodology): “The proposed feature-space discriminative linear regression or fDLR is similar to LIN [4, 7, 12] with two differences.”
Learning hidden unit contributions for unsupervised speaker adaptation of neural network acoustic models (2014, background): “Explicit adaptation to speaker or acoustic characteristics can further improve accuracy [12, 13, 14, 15, 16].”