Who Needs Words? Lexicon-Free Speech Recognition

By Tatiana Likhomanenko, Gabriel Synnaeve, Ronan Collobert. 2019

27 citations recorded by the source corpus.

Lexicon-free speech recognition naturally deals with the problem of out-of-vocabulary (OOV) words. In this paper, we show that character-based language models (LM) can perform as well as word-based LMs for speech recognition, in word error rates (WER), even without restricting the decoding to a lexicon. We study character-based LMs and show that convolutional LMs can effectively leverage large (character) contexts, which is key for good speech recognition performance downstream. We specifically show that the lexicon-free decoding performance (WER) on utterances with OOV words using character-based LMs is better than lexicon-based decoding, both with character or word-based LMs.

Subfields: Artificial Intelligence, Natural Language Processing

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

In Natural Language Processing

Citations in the corpus

End-to-end ASR: from Supervised to Semi-Supervised Learning with Modern Architectures (2019, positive): “In our experiments, we use lexicon-based and lexicon-free beam-search decoders following (Collobert et al., 2016; Likhomanenko et al., 2019) with either n-gram or GCNN LMs.”
vq-wav2vec: Self-Supervised Learning of Discrete Speech Representations (2019, methodology): “We consider a 4-gram KenLM language model (Heaﬁeld et al., 2013) and a character based convolutional language model (Likhomanenko et al., 2019) and tune the models with the same protocol as Schneider et al. (2019).”
wav2vec: Unsupervised Pre-training for Speech Recognition (2019, neutral): “, 2019), and a character based convolutional language model (Likhomanenko et al., 2019).”
Scaling Speech Technology to 1, 000+ Languages (2023, methodology): “For decoding with word-level LMs, we use the lexicon-based decoder of Collobert et al. [2016], Pratap et al. [2019] and for character-level LMs, we use the lexicon-free beam-search decoder of Likhomanenko et al. [2019].”
An open-source end-to-end ASR system for Brazilian Portuguese using DNNs built from newly assembled corpora (2020, methodology): “This approach, used by [50], does not take into account that word-level models are constrained to a fixed-size lexicon, while character-based models have virtually an infinite vocabulary size.”
slimIPL: Language-Model-Free Iterative Pseudo-Labeling (2020, methodology): “We use open-sourced word-level LMs trained on the LibriSpeech LM corpus: 4-gram [45] and Transformer [8] LMs.”
Joint Masked CPC And CTC Training For ASR (2020, methodology): “Besides reporting word error rate (WER) without an LM, we also perform a one-pass beam-search decoder with a 4-gram word-level LM [25] and further the beam rescoring with a strong word-level Transformer LM [26].”
Word-level Speech Recognition with a Dynamic Lexicon (2019, methodology): “Unlike lexicon-free approaches [22, 25, 33] our model requires a lexicon.”
End-to-End Speech Recognition: A Survey (2023, background): “Thus, they achieve competitive performance to that of RNN-LMs [292], even with the ﬁnite history consisting of short tokens such as characters [294].”
Speech Recognition (2018, unknown): “, 2019), and a character based convolutional language model (Likhomanenko et al., 2019).”
Word-Level Speech Recognition With a Letter to Word Encoder (2019, methodology): “Unlike lexicon-free approaches (Likhomanenko et al., 2019; Maas et al., 2015; Zeyer et al., 2018) our model requires a lexicon.”