Enhancing the TED-LIUM Corpus with Selected Data for Language Modeling and More TED Talks

By Anthony Rousseau, Paul Deléglise, Yannick Estève. 2014

304 citations recorded by the source corpus.

In this paper, we present improvements made to the TED-LIUM corpus we released in 2012. These enhancements fall into two categories. First, we describe how we filtered publicly available monolingual data and used it to estimate well-suited language models (LMs), using open-source tools. Then, we describe the process of selection we applied to new acoustic data from TED talks, providing additions to our previously released corpus. Finally, we report some experiments we made around these improvements.

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

In Natural Language Processing

In Speech & Audio

Citations in the corpus

TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation (2018, methodology): “These source data are described in [11].”
The CAPIO 2017 Conversational Speech Recognition System (2017, methodology): “To see how well our approach can be generalized to nontelephony environments, we have trained separate systems using the TED-LIUM release 2 [39, 40] and LibriSpeech [41] data.”
A Comparison of Transformer and LSTM Encoder Decoder Models for ASR (2019, methodology): “The experiments were performed on LibriSpeech (1000h) [59], Switchboard (300h) [60], and TED-LIUM release 2 (200h) [61].”
FINDINGS OF THE IWSLT 2020 EVALUATION CAMPAIGN (2020, unknown): “…TED corpus • How2 (Sanabria et al., 2018) 8 • LibriVoxDeEn (Beilharz and Sun, 2019) • Europarl-ST (Iranzo-S´anchez et al., 2020) • TED LIUM v2 (Rousseau et al., 2014) and v3 (Hernandez et al., 2018) • all the data provided by WMT 2019 • OpenSubtitles 2018 (Lison et al., 2018) • Augmented…”
Self-Attentional Acoustic Models (2018, methodology): “We focus our experiments on the TEDLIUM corpus [17], a widely used corpus of 200h of recorded TED talks, with the development split used as validation data.”
Data Augmentation for ASR Using TTS Via a Discrete Representation (2021, methodology): “After data augmentation with the TTS, we trained and evaluated the ASR models with LibriSpeech and TED-LIUM release-2 [34].”
Unspeech: Unsupervised Speech Context Embeddings (2018, background): “TEDLIUM V2 [26] has a comparatively small number of speakers, especially in the development and test set of the corpus.”
The Rwth Asr System for Ted-Lium Release 2: Improving Hybrid Hmm With Specaugment (2020, methodology): “In this work, we describe a complete training pipeline to build a state-of-the-art hybrid HMM-based ASR system on the 2nd release of the TED-LIUM corpus [14] (TED-LIUMv2).”
Semantic Mask for Transformer based End-to-End Speech Recognition (2019, methodology): “To verify the generalization of the semantic mask, we further conduct experiments on TedLium2 [23] dataset, which is extracted from TED talks.”
Open Source Toolkit for Speech to Text Translation (2018, methodology): “For training the speech recognition systems we use the version 2 of the Tedlium Corpus (Rousseau et al., 2014).”
A Time-Restricted Self-Attention Layer for ASR (2018, methodology): “Experiments are done on four corpora: Wall street Journal [13], TED-LIUM [14], Switchboard [15] and AMI.”