Speech recognition and keyword spotting for low-resource languages: Babel project research at CUED

By Mark John Francis Gales, K. Knill, Anton Ragni, Shakti Prasad Rath. 2014

206 citations recorded by the source corpus.

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

FLEURS: FEW-Shot Learning Evaluation of Universal Representations of Speech (2022, methodology): “We initialize fine-tuning from a 600M parameter wav2vec-BERT [22] model, which has been pre-trained on 429k hours unlabeled speech data in 51 languages pooling from VoxPopuli [9], MLS [8], CommonVoice [12] and BABEL [13].”
MAESTRO: Matched Speech Text Representations through Modality Matching (2022, methodology): “Multilingual ASR: Following [11], we use 429k hours of public unlabeled speech corpora: VoxPopuli [33], CommonVoice [34], MLS [35] and BABEL [36] to pre-train multilingual ASR models.”
mSLAM: Massively multilingual joint pre-training for speech and text (2022, methodology): “1 release of Common Voice (Ardila et al., 2019), 50k hours of read books data in eight European languages from Multilingual LibriSpeech (Pratap et al., 2020) and 1k hours of telephonic conversation data spanning 17 African and Asian languages from BABEL (Gales et al., 2014).”
XLS-R: Self-supervised Cross-lingual Speech Representation Learning at Scale (2021, methodology): “BABEL (BBL) is a multilingual corpus of conversational telephone speech of about 1K hours of data in 17 African and Asian languages [22].”
Common Voice: A Massively-Multilingual Speech Corpus (2019, background): “Even though the Babel corpus contains high-quality data from 22 minority languages, it is not released under an open license.”
Unsupervised Cross-lingual Representation Learning for Speech Recognition (2020, methodology): “We evaluate XLSR on 5 languages of the BABEL benchmark [29] which is conversational telephone data and ten languages of CommonVoice [30], a corpus of read speech (§ 3).”
Scaling Speech Technology to 1, 000+ Languages (2023, methodology): “…hours of YouTube content [Valk and Alumäe, 2020] • BABEL (BBL): 17 African and Asian languages totaling about 1K hours of conversational telephone data [Gales et al., 2014] • VoxPopuli (VP): 371K hours of unlabeled speech data in 23 languages derived from European Parliament event recordings [Wang…”
Data Augmentation for Deep Neural Network Acoustic Modeling (2015, background): “Two data augmentation approaches, vocal tract length perturbation (VTLP) and stochasticfeature mapping (SFM), are investigated for both deep neural networks (DNNs) and convolutional neural networks (CNNs).”
Automatic speaker verification spoofing and deepfake detection using wav2vec 2.0 and data augmentation (2022, methodology): “We have hence explored the wav2vec 2.0 XLS-R (0.3B) model1 [13] in our work. wav2vec 2.0 XLS-R is a large-scale cross-lingually pre-trained model trained on diverse corpora including VoxPopuli data (VP-400K) [41], the multilingual Librispeech corpus (MLS) [42], CommonVoice (CV) [43], VoxLingua107 (V”
An Investigation of Deep Neural Networks for Multilingual Speech Recognition Training and Adaptation (2017, methodology): “Speech recognition systems built with multilingual deep neural networks (DNNs) have been shown to provide consistent advantages especially for lowresourced languages [1, 2, 3, 4].”
Knowledge distillation across ensembles of multilingual models for low-resource languages (2017, methodology): “Multilingual models are trained using Multilingual (ML) features and have been the key for achieving good performance in ASR and keyword search (KWS) tasks in the Babel program [17, 18, 19, 20, 21, 22].”
Massively Multilingual Adversarial Speech Recognition (2019, background): “Prior work in multilingual and cross-lingual speech recognition has been restricted to a small handful of the world’s most-spoken languages, relying on multilingual corpora such as GlobalPhone (Schultz, 2002), the IARPA Babel corpora (Gales et al., 2014), or the VoxForge1 corpora.”
Google USM: Scaling Automatic Speech Recognition Beyond 100 Languages (2023, background): “1 release of Common Voice [75], 50k hours of read books data in eight European languages from Multilingual LibriSpeech [76] and 1k hours of telephonic conversation data spanning 17 African and Asian languages from BABEL [77].”
Self-Supervised Speech Representation Learning: A Review (2022, unknown): “These corpora include Common Voice (CV) [214], Multilingual LibriSpeech (MLS) [215], VoxPopuli (VP) [216], and BABEL (BBL) [217].”
XTREME-S: Evaluating Cross-lingual Speech Representations (2022, background): “Approaches like XLS-R [4, 5], for example, have shown particularly strong results on several tasks, including ASR on BABEL and multilingual LibriSpeech, and AST on CoVoST-2.”
Leveraging unsupervised and weakly-supervised data to improve direct speech-to-speech translation (2022, methodology): “[35] M. J. Gales, K. M. Knill, A. Ragni, and S. P. Rath, “Speech recognition and keyword spotting for low-resource languages: Babel project research at cued,” in Proc.”
Low-Resource Speech Recognition and Keyword-Spotting (2017, methodology): “This section brieﬂy describes some of the approaches adopted for implementing the lexicon, acoustic and language models in those low-resource scenarios at CUED to support the evaluation systems developed over the Babel programme [11,26,25].”