Grapheme-to-phoneme conversion using Long Short-Term Memory recurrent neural networks

By Kanishka Rao, Fuchun Peng, Hasim Sak, Franccoise Beaufays. 2015

202 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Natural Language Processing, 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 Artificial Intelligence

1. Tao Qin
2. Tie-Yan Liu
3. Xu Tan
4. Jinyu Li
5. Dong Yu
6. Tara N. Sainath
7. Quoc V. Le
8. Yoshua Bengio

In Natural Language Processing

1. Tie-Yan Liu
2. Tao Qin
3. Dong Yu
4. Yoshua Bengio
5. Xu Tan
6. Quoc V. Le
7. Ronan Collobert
8. Jinyu Li

In Speech & Audio

1. Jinyu Li
2. Xu Tan
3. Tao Qin
4. Tara N. Sainath
5. Tie-Yan Liu
6. Rohit Prabhavalkar
7. Sheng Zhao
8. Dong Yu

Citations in the corpus

• Deep Voice: Real-time Neural Text-to-Speech (2017, methodology): “Previous work uses neural networks as substitutes for several TTS system components, including grapheme-to-phoneme conversion models (Rao et al., 2015; Yao & Zweig, 2015), phoneme duration prediction models (Zen & Sak, 2015), fundamental frequency prediction models (Pascual & Bonafonte, 2016;…”
• A Survey on Neural Speech Synthesis (2021, methodology): “Thus, the G2P conversion for English is mainly responsible to generate the pronunciations of out-of-vocabulary words [41, 24, 403, 283, 33, 320].”
• Sequence-to-sequence neural net models for grapheme-to-phoneme conversion (2015, methodology): “It is possible to improve performances by combining multiple systems and methods [32, 33], we have chosen not to engage in building hybrid models.”
• Exploring architectures, data and units for streaming end-to-end speech recognition with RNN-transducer (2017, methodology): “The pronunciation model, P ( φ | W ) , is typically built from pronunciation dictionaries curated by expert human linguists, with back-off to a grapheme-to-phoneme (G2P) model [6] for out of dictionary words.”
• Token-Level Ensemble Distillation for Grapheme-to-Phoneme Conversion (2019, background): “In WER calculation, considering the multiple pronunciations, word error is counted only when the output differs from all the references, following [2, 3, 18, 28].”
• Improving Attention Based Sequence-to-Sequence Models for End-to-End English Conversational Speech Recognition (2018, methodology): “The performance of speech recognition has been improved dramatically since deep neural networks (DNNs) were applied to its main components such as acoustic model [1–6], language model [7, 8], pronunciation model [9], etc.”
• Listen, attend and spell: A neural network for large vocabulary conversational speech recognition (2015, background): “DNNs have also produced significant gains in pronunciation models that map words to phoneme sequences [7, 8].”
• RNN Approaches to Text Normalization: A Challenge (2016, methodology): “For the channel LSTM we used a bidirectional sequence-to-sequence model similar to that reported in (Rao et al., 2015) in two configurations: one with two forward and two backward hidden layers, hence-forth the shallow model; and one with three forward and three backward hidden layers, henceforth…”
• Transliterated Mobile Keyboard Input via Weighted Finite-State Transducers (2017, methodology): “, 2014) and neural sequence models (Rao et al., 2015).”
• Word-Level Speech Recognition With a Letter to Word Encoder (2019, background): “Traditional speech recognizers also have adaptable lexicons, which they achieve by predicting phonemes and requiring a model for grapheme-to-phoneme conversion (Bisani & Ney, 2008; Rao et al., 2015).”
• Representation Mixing for TTS Synthesis (2018, unknown): “Alternate approaches such as grapheme to phoneme methods [10] will also be unable to resolve this problem.”
• A bi-directional LSTM approach for polyphone disambiguation in Mandarin Chinese (2016, methodology): “In the G2P conversion task [1, 2, 3, 4, 5, 6, 7], a neural network accepts a sequence of characters and outputs a sequence of pronunciations.”
• Personalized speech recognition on mobile devices (2016, methodology): “The LSTM-G2P performs better in terms of word accuracy compared to traditional joint-sequence models represented as finite state transducers (FSTs) (a detailed comparison can be found in [25]).”
• Word-level Speech Recognition with a Dynamic Lexicon (2019, background): “Traditional speech recognizers also have adaptable lexicons, which they achieve by predicting phonemes and requiring a model for grapheme-to-phoneme conversion [7, 29].”
• Chinese syllable-to-character conversion with recurrent neural network based supervised sequence labelling (2015, methodology): “For labelling unsegmented sequence data with RNNs, the Connectionist Temporal Classification (CTC) algorithm have been successfully used in many tasks, such as the handwriting recognition [9], automatic speech recognition [14], [8] and grapheme-to-phoneme conversion [10].”
• The SIGMORPHON 2020 Shared Task on Multilingual Grapheme-to-Phoneme Conversion (2020, background): “…2005, Bisani and Ney 2008, Wu et al. 2014, Novak et al. 2016) or discriminative models based on conditional random fields (Lehnen et al. 2013), recurrent neural networks (e.g., Rao et al. 2015, Yao and Zweig 2015, van Esch et al. 2016, Lee et al. 2020) or transformers (Yolchuyeva et al. 2019).”
• Polyphone Disambiguation for Mandarin Chinese Using Conditional Neural Network with Multi-level Embedding Features (2019, methodology): “G2P typically generates a sequence of phones from a sequence of characters or graphemes [7].”