Speaker-independent phone recognition using hidden Markov models

By Kai-Fu Lee, Hsiao-Wuen Hon. 1989

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

Speech recognition with deep recurrent neural networks (2013, neutral): “As with CTC, each distribution covers the K phonemes plus ∅; the network now ‘decides’ what to output depending both on where it is in the input sequence and the outputs it has already emitted.”
Predicting Parameters in Deep Learning (2013, methodology): “the phones in each utterance using a bigram language model, and confusions between certain sets of phones were ignored as described in [19].”
Sequence Transduction with Recurrent Neural Networks (2012, methodology): “This may be because the TIMIT transcriptions are too small a training set for the prediction network: around 150K labels, as opposed to the millions of words typically used to train language models.”
Fixed-point feedforward deep neural network design using weights +1, 0, and −1 (2014, methodology): “For evaluation, the original 61 phones are folded into standard 39 classes as proposed in [18].”
Deep Rewiring: Training very sparse deep networks (2017, methodology): “There are 61 different phonemes annotated in the TIMIT dataset, to report an error rate that is comparable to the literature we performed a standard grouping of the phonemes to generate 39 output classes (Lee & Hon, 1989; Graves et al., 2013; Greff et al., 2017).”
TERA: Self-Supervised Learning of Transformer Encoder Representation for Speech (2020, methodology): “Following the conventional settings [54], ASR modelling on TIMIT is based on 48 phoneme classes, while accuracy is measured after mapping the prediction to a smaller set of 39 phoneme classes.”
Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks (2013, neutral): “The 61 hand labeled phonetic symbols are mapped to 39 phonemes with an additional garbage class, as presented in [18].”
Analyzing Hidden Representations in End-to-End Automatic Speech Recognition Systems (2017, methodology): “We also experimented with the reduced set of 48 phones used by [35].”
Analysis of CNN-based speech recognition system using raw speech as input (2015, neutral): “The 61 hand labeled phonetic symbols are mapped to 39 phonemes with an additional garbage class, as presented in [21].”
The DIRHA-ENGLISH corpus and related tasks for distant-speech recognition in domestic environments (2015, methodology): “In the evaluation phase, the standard Kaldi recipe (based on Sclite ) maps the original 48 English phones into a reduced set of 39 units, as originally done in [34].”
SafetyNets: Verifiable Execution of Deep Neural Networks on an Untrusted Cloud (2017, methodology): “During testing, all labels are mapped to 39 classes [11] for evaluation.”
Fast Adaptation of Deep Neural Network Based on Discriminant Codes for Speech Recognition (2014, methodology): “After decoding, these 61 phone classes are mapped to a set of 39 classes for the final scoring purpose as in [32].”
Phone Recognition on the TIMIT Database (2012, methodology): “For more information visit www.intechopen.com Open access books available Countries delivered to Contributors from top 500 universities International authors and editors Our authors are among the most cited scientists Downloads We are IntechOpen, the world’s leading publisher of”
Towards Lower Error Rates in Phoneme Recognition (2004, methodology): “It is very similar to the CMU/MIT phoneme set [2], but closures were merged with burst instead of with silence (bcl b → b).”
On internal language representations in deep learning: an analysis of machine translation and speech recognition (2018, result): “Finally, Figure 5-7 reports experiments with a reduced set of 48 phones [195], exhibiting a similar trend to the other label sets.”
Convolutional Neural Networks-based continuous speech recognition using raw speech signal (2015, unknown): “The 61 hand labeled phonetic symbols are mapped to 39 phonemes with an additional garbage class, as presented in [22].”
Deep Belief Networks using discriminative features for phone recognition (2011, methodology): “After decoding, the 61 phone classes were mapped to a set of 39 classes as in [13] for scoring.”