SWITCHBOARD: telephone speech corpus for research and development

By J. Godfrey, E. Holliman, J. McDaniel. 1992

Subfields: Natural Language Processing, Evaluation & Benchmarks

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Natural Language Processing

In Evaluation & Benchmarks

Citations in the corpus

Benchmarking Deep Reinforcement Learning for Continuous Control (2016, background): “Examples include MNIST (LeCun et al., 1998), Caltech101 (Fei-Fei et al., 2006), CIFAR (Krizhevsky & Hinton, 2009), ImageNet (Deng et al., 2009), PASCAL VOC (Everingham et al., 2010), BSDS500 (Martin et al., 2001), SWITCHBOARD (Godfrey et al., 1992), TIMIT (Garofolo et al., 1993), Aurora (Hirsch & Pe”
Earnings-21: A Practical Benchmark for ASR in the Wild (2021, methodology): “Whereas the best reported word error rate (WER) results on LibriSpeech (around 1.4% [11]) or Switchboard (around 5% [12]) are encouraging, in reality, most commercially available systems are much closer to 15-20% [10] when transcribing user-provided recordings, making ASR a problem that is far from ”
CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for Unsegmented Recordings (2020, neutral): “In parallel to research on acoustic robustness, research on conversational speech recognition has also made great progress, as illustrated by the recent announcements of super-human performance [22, 23] achieved on the Switchboard telephone conversation task [24] and by the ASpIRE challenge [25].”
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition (2019, background): “On Switchboard 300h (LDC97S62) [22], we obtain 7.2% WER on the Switchboard portion of the Hub5’00 (LDC2002S09, LDC2003T02) test set, and 14.6% on the Call-Home portion, without using an LM.”
Toward Domain-Invariant Speech Recognition via Large Scale Training (2018, methodology): “In [7], transfer learning is used to adapt a model trained on switchboard [15] to improve performance on domains with smaller training sets like WSJ [16] and AMI [17].”
The Ubuntu Dialogue Corpus: A Large Dataset for Research in Unstructured Multi-Turn Dialogue Systems (2015, methodology): “Switchboard [7] Human-human Various 2,400 — 3,000,000 Telephone conversations spoken on pre-specified topics DSTC1 [30] Human-computer State 15,000 210,000 Bus ride information spoken tracking system DSTC2 [9] Human-computer State 3,000 24,000 — Restaurant booking spoken tracking system DSTC3 [8] Hu”
A Survey of Available Corpora For Building Data-Driven Dialogue Systems: The Journal Version (2015, background): “Perhaps one of the most influential spoken corpora is the Switchboard dataset (Godfrey et al., 1992).”
Gender and Dialect Bias in YouTube’s Automatic Captions (2017, background): “Switchboard (Godfrey et al., 1992) undersamples women, Southerns and non-college-educated speakers.”
Audiobox: Unified Audio Generation with Natural Language Prompts (2023, background): “First, we annotate a set of 1,946 recordings sampled from diverse sources, including LibriTTS (Zen et al., 2019), Common Voice (Ardila et al., 2019), Switchboard (Godfrey et al., 1992), Fisher (Cieri, Christopher, et al. , 2004,2005a,,), Spotify (Clifton et al., 2020), AudioSet (Gemmeke et al., 2017”
Voicebox: Text-Guided Multilingual Universal Speech Generation at Scale (2023, background): “Hence, the current models may not transfer well to conversational speech [15], which is more casual and contains more non-verbal sounds such as laughing and back-channeling (e.g., um-hmm).”
VoxCeleb: A Large-Scale Speaker Identification Dataset (2017, background): “ELSDSR [26] Clean Speech 22 198 MIT Mobile [21] Mobile Devices - 88 7,884 SWB [27] Telephony - 3,114 33,039 POLYCOST [17] Telephony - 133 1,285 ‡ ICSI Meeting Corpus [23] Meetings - 53 922 Forensic Comparison [22] Telephony 552 1,264 ANDOSL [18] Clean speech - 204 33,900 TIMIT [28] † Clean speech - ”
Achieving Human Parity in Conversational Speech Recognition (2016, background): “The 500-best lists from the lattices generated with the ResNet CNN system had an oracle (lowest achievable) WER of 2.7% on the Switchboard portion of the NIST 2000 evaluation set, and an oracle WER of 4.9% on the CallHome portion.”
Lessons from Building Acoustic Models with a Million Hours of Speech (2019, background): “Over the decades, corpora have grown from a few tens of hours of speech (TIDIGITS [5], TIMIT [6], WSJ [7]), to a few hundred hours (Switchboard [8]), to a few thousand hours of speech (Fisher corpus [9]).”
A Comparative Study on Transformer vs RNN in Speech Applications (2019, background): “ev JSUT [25] ja 10 read (our split) LibriSpeech [34] en 960 clean/noisy read dev clean / dev other / test clean / test other REVERB [35] (*) en 124 far-ﬁeld multi-ch read et near / et far SWITCHBOARD [36] en 260 telephone conversational (eval2000) callhm / swbd TED-LIUM2 [37] en 118 spontaneous dev ”
DailyDialog: A Manually Labelled Multi-turn Dialogue Dataset (2017, methodology): “This makes DailyDialog distinguished from existing dialog datasets such as Switchboard (Godfrey et al., 1992) and OpenSubtitles (Jörg Tiedemann, 2009), which often have 150+ and 1,000+ speaker turns in one “conversation”.”
AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation, Recognition and Speaker Diarization in Conference Scenario (2021, methodology): “For ASR with conversation speech, switchboard [11], collected by Texas Instruments, is a dataset of telephone communication with about 2,500 conversations by 500 native speakers.”
Redundancy and reduction: Speakers manage syntactic information density (2010, methodology): “The data comes from the Penn Treebank (release 3, Marcus, Santorini, Marcinkiewicz, & Taylor, 1999) subset of the Switchboard corpus of telephone dialogues (Godfrey et al., 1992).”