Lightly supervised and unsupervised acoustic model training

By Lori Faith Lamel, Jean-Luc Gauvain, Gilles Adda. 2002

307 citations recorded by the source corpus.

Subfields: Artificial Intelligence, Speech & Audio, Signal 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 Speech & Audio

In Signal Processing

Citations in the corpus

Lessons from Building Acoustic Models with a Million Hours of Speech (2019, background): “There is evidence that even unﬁltered data can lead to signiﬁcant SSL improvements ([17], [37]).”
Unsupervised spoken keyword spotting via segmental DTW on Gaussian posteriorgrams (2009, background): “The work of Lamel et al. for example showed the potential for training with lightly annotated data [ 4 ].”
slimIPL: Language-Model-Free Iterative Pseudo-Labeling (2020, methodology): “ and web-scale bitext mining [44]; • automatic speech recognition (ASR): representation learning [45, 2, 46, 47], local prior matching [4], adversarial training [48], back-translation [49] and others [50, 51, 52, 53, 54]. Below, we give an overview of the approaches in ASR that are most recent and r”
Analysis of low-resource acoustic model self-training (2009, background): “Lamel et al. (2002) demonstrated that ten minutes of transcribed broadcast audio are sufficient to achieve a 33% relative reduction in WER when using 135 hours of unlabeled audio and a strong broadcast news language model (LM).”
Semi-supervised maximum mutual information training of deep neural network acoustic models (2015, methodology): “Therefore sequence-discriminative self-training methods do not work well without some form of confidence-based filtering, as used in [20, 21, 17, 22].”
Semi-supervised Training for End-to-end Models via Weak Distillation (2019, methodology): “Confidence scores from an existing ASR system is commonly used to select unsupervised data for training with more data [11–15].”
Unsupervised versus supervised training of acoustic models (2008, methodology): “[5, 8] reports unsupervised training experiments on 153 hours of broadcast news (BN) data.”
Large scale deep neural network acoustic modeling with semi-supervised training data for YouTube video transcription (2013, methodology): “There has been previous work on using closed captions for training [11, 12].”
Speech recognition and keyword spotting for low-resource languages: Babel project research at CUED (2014, background): “In this scenario it is possible to use semi-supervised training [16, 17, 18].”
Unsupervised training of acoustic models for large vocabulary continuous speech recognition (2005, background): “In [4], [5], and [6], closed captions are used instead of confidence measures to filter the recognized transcriptions.”
Semi-Supervision in ASR: Sequential MixMatch and Factorized TTS-Based Augmentation (2021, methodology): “The traditional method of training on untranscribed speech uses predictions from a model trained on supervised data to generate pseudo labels for the untranscribed data [5,6].”
A factor automaton approach for the forced alignment of long speech recordings (2009, background): “Index Terms— finite state transducers, speech alignment, speech recognition.”
A big data approach to acoustic model training corpus selection (2014, methodology): “Rather than relying on small supervised training sets, one can instead exploit the large amount of audio data that is processed by systems like Voice Search, together with semi-supervised techniques to construct large training sets with automatically derived transcripts [4, 5, 6, 7, 8].”
Unsupervised training and directed manual transcription for LVCSR (2010, methodology): “These approximate transcriptions can be used with lightly-supervised training (Lamel et al., 2002; Chan and Woodland, 2004).”
Self-Supervised Speech Representation Learning: A Review (2022, methodology): “relies on a supervised teacher model to label a large volume of speech-only data, which is then used to augment the initial labeled data to train a student model [4], [5], [6], [61].”
Lightly supervised recognition for automatic alignment of large coherent speech recordings (2010, methodology): “The proposed approach can be described as lightly supervised recognition [6] plus sentence selection (henceforth called LightlySupervised ).”
Current trends in multilingual speech processing (2011, methodology): “There has also been increasing effort in the use of lightly supervised and unsupervised training methods for ASR (Lamel et al 2002), which has more recently been applied to TTS (Yamagishi et al 2009).”
Investigations on large-scale lightly-supervised training for statistical machine translation. (2008, methodology): “These resources used to build the initial SMT system are usually not considered as supervision in the framework of unsupervised training applied to additional data[7, 8].”