Uncertainty weighting and propagation in DNN–HMM-based speech recognition

Nóvoa, José; Fredes, Josué; Poblete, Víctor; Yoma, Néstor Becerra

doi:10.1016/j.csl.2017.06.005

Cited by 32 publications

(8 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…It should be noted that the Viterbi path is the most probable state sequence based on the state transition probabilities and the probability of the observed values fitting the emissions distribution of a given state. While the Viterbi path is a useful tool for visualizing the state occurrences through time, there is, however, uncertainty associated with the state classification that has been addressed in studies elsewhere (Hernando et al 2005;Novoa et al 2018).…”

Section: Resultsmentioning

confidence: 99%

How Wet and Dry Spells Evolve across the Conterminous United States Based on 555 Years of Paleoclimate Data

Lall

Cook

2018

Journal of Climate

View full text Add to dashboard Cite

Evolving patterns of droughts and wet spells in the conterminous United States (CONUS) are examined over 555 years using a tree-ring-based paleoclimate reconstruction of the modified Palmer drought severity index (PDSI). A hidden Markov model is used as an unsupervised method of classifying climate states and quantifying the temporal evolution from one state to another. Modeling temporal variability in spatial patterns of drought and wet spells provides the ability to objectively assess and simulate historical persistence and recurrence of similar patterns. The Viterbi algorithm reveals the probable sequence of states through time, enabling an examination of temporal and spatial features and associated large-scale climate forcing. Distinct patterns of sea surface temperature that are known to enhance or inhibit rainfall are associated with some states. Using the current CONUS PDSI field the model can be used to simulate the space–time PDSI pattern over the next few years, or unconditional simulations can be used to derive estimates of spatially concurrent PDSI patterns and their persistence and intensity across the CONUS.

show abstract

Section: Resultsmentioning

confidence: 99%

How Wet and Dry Spells Evolve across the Conterminous United States Based on 555 Years of Paleoclimate Data

Lall

Cook

2018

Journal of Climate

View full text Add to dashboard Cite

show abstract

“…Deep learning produces successful results for many problems that cause problems in the advancement of artificial intelligence methods. It is used in many areas such as image recognition [17][18][19], speech processing [20,21], and gene mutation prediction [22]. It is thought that DNN will make faster progress thanks to the new neural network structures proposed with an increase in computation capacity and amount of data [16].…”

Section: Deep Neural Network (Dnn)mentioning

confidence: 99%

Brain Tumor Classification Using Deep Neural Network

Çınarer¹,

Arslan²,

Yurttakal³

2020

Adv. sci. technol. eng. syst. j.

View full text Add to dashboard Cite

Brain tumors are a type of tumor with a high mortality rate. Since multifocal-looking tumors in the brain can resemble multicentric gliomas or gliomatosis, accurate detection of the tumor is required during the treatment process. The similarity of neurological and radiological findings also complicates the classification of these tumors. Fast and accurate classification is important for brain tumors. Computer aided diagnostic systems and deep neural network architectures can be used in the diagnosis of multicentric gliomas and multiple lesions. In this study, the Deep Neural Network classification model with Synthetic Minority Over-sampling Technique pre-processing was used on the Visually Accessible Rembrandt Images dataset. The proposed model for the classification of brain tumors consists of 1319 trainable parameters and the proposed method has achieved 95.0% accuracy rate. Precision, Recall, F1-measure values are 95.4%, 95.0% and 94.9% respectively. The proposed decision support system can be used to give an idea to doctors in the detection of glioma type tumors.

show abstract

“…The use of conventional hidden Markov models (HMMs) and deep neural networks (DNNs) of automatic speech recognition (ASR) systems in the preparation of a lexicon, acoustic models, and language models results in complications [1]. These approaches also require linguistic resources, such as a pronunciation dictionary, tokenization, and phonetic context dependencies [2]. In contrast, end-to-end ASR has grown to be a popular alternative to simplify the conventional ASR model building process.…”

Section: Introductionmentioning

confidence: 99%

“…The Amharic language mainly consists of seven vowels [18], namely ኧ[ә], ኡ[u], ኢ[i], ኣ[a], ኤ[e], እ [1], ኦ [o]. This language has 32 consonants [18,22] that are categorized based on their articulation stops (14), fricatives (8), affricatives (3), nasals (3), liquids (2), and glides (2). These consonants are indicated in Appendix B with their corresponding International Phonetic Alphabet (IPA) representations.…”

Section: Introductionmentioning

confidence: 99%

Improving Amharic Speech Recognition System Using Connectionist Temporal Classification with Attention Model and Phoneme-Based Byte-Pair-Encodings

Emiru

Xiong

et al. 2021

Information

View full text Add to dashboard Cite

Out-of-vocabulary (OOV) words are the most challenging problem in automatic speech recognition (ASR), especially for morphologically rich languages. Most end-to-end speech recognition systems are performed at word and character levels of a language. Amharic is a poorly resourced but morphologically rich language. This paper proposes hybrid connectionist temporal classification with attention end-to-end architecture and a syllabification algorithm for Amharic automatic speech recognition system (AASR) using its phoneme-based subword units. This algorithm helps to insert the epithetic vowel እ[ɨ], which is not included in our Grapheme-to-Phoneme (G2P) conversion algorithm developed using consonant–vowel (CV) representations of Amharic graphemes. The proposed end-to-end model was trained in various Amharic subwords, namely characters, phonemes, character-based subwords, and phoneme-based subwords generated by the byte-pair-encoding (BPE) segmentation algorithm. Experimental results showed that context-dependent phoneme-based subwords tend to result in more accurate speech recognition systems than the character-based, phoneme-based, and character-based subword counterparts. Further improvement was also obtained in proposed phoneme-based subwords with the syllabification algorithm and SpecAugment data augmentation technique. The word error rate (WER) reduction was 18.38% compared to character-based acoustic modeling with the word-based recurrent neural network language modeling (RNNLM) baseline. These phoneme-based subword models are also useful to improve machine and speech translation tasks.

show abstract

Uncertainty weighting and propagation in DNN–HMM-based speech recognition

Cited by 32 publications

References 9 publications

How Wet and Dry Spells Evolve across the Conterminous United States Based on 555 Years of Paleoclimate Data

How Wet and Dry Spells Evolve across the Conterminous United States Based on 555 Years of Paleoclimate Data

Brain Tumor Classification Using Deep Neural Network

Improving Amharic Speech Recognition System Using Connectionist Temporal Classification with Attention Model and Phoneme-Based Byte-Pair-Encodings

Contact Info

Product

Resources

About