The Fisher Corpus: a Resource for the Next Generations of Speech-to-Text

By C. Cieri, David Miller, Kevin Walker. 2004

686 citations recorded by the source corpus.

This paper describes, within the context of the DARPA EARS program, the design and implementation of the Fisher protocol for collecting conversational telephone speech which has yielded more than 16,000 English conversations. It also discusses the Quick Transcription specification that allowed 2000 hours of Fisher audio to be transcribed in less than one year. Fisher data is already in use within the DARPA EARS programs and will be published via the Linguistic Data Consortium for general use beginning in 2004.

Subfields: Artificial Intelligence, Evaluation & Benchmarks, 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 Evaluation & Benchmarks

In Speech & Audio

Citations in the corpus

SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition (2019, methodology): “Upon shallow fusion with an LM trained on the combined transcript of the Switch-board and Fisher (LDC200 { 4,5 } T19) [23] corpora, we obtain 6.8%/14.1% WER on the Switchboard/Callhome portion.”
Deep Speech 2 : End-to-End Speech Recognition in English and Mandarin (2015, background): “The Wall Street Journal (WSJ), Switchboard and Fisher [13] corpora are all published by the Linguistic Data Consortium.”
Deep Speech: Scaling up end-to-end speech recognition (2014, methodology): “We evaluate our system trained on only the 300 hour Switchboard conversational telephone speech dataset and trained on both Switchboard (SWB) and Fisher (FSH) [3], a 2000 hour corpus collected in a similar manner as Switchboard.”
Generative Spoken Dialogue Language Modeling (2022, methodology): “Our approach is based on the availability of a dataset constructed along the Fisher Telephone conversation collection protocol (Cieri et al., 2004) where each conversation involves two speakers, and each speaker is recorded in a separate audio channel while having a very casual conversation.”
Achieving Human Parity in Conversational Speech Recognition (2016, background): “Past work [41] reports inter-transcriber error rates for data taken from the later RT03 test set (which contains Switchboard and Fisher, but no CallHome data).”
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]).”
Generating High-Quality and Informative Conversation Responses with Sequence-to-Sequence Models (2017, neutral): “The Fisher corpus (David et al., 2004).”
CVSS Corpus and Massively Multilingual Speech-to-Speech Translation (2022, neutral): “, 1993), Fisher (Cieri et al., 2004), and Mixer 4 and 5 (Cieri et al.”
JHU ASpIRE system: Robust LVCSR with TDNNS, iVector adaptation and RNN-LMS (2015, methodology): “Speech from the English portion of the Fisher corpus [9] (LDC2004S13, LDC2005S13) was used to train the acoustic models.”
A Comparison of Techniques for Language Model Integration in Encoder-Decoder Speech Recognition (2018, methodology): “For external LM training, we combine the Switchboard training set with the Fisher corpus (LDC200 { 4,5 } T19) [21].”
Building DNN acoustic models for large vocabulary speech recognition (2014, methodology): “To maximize the amount of training data for a conversational speech transcription task, we combine the Switchboard corpus with the larger Fisher corpus [69].”
Multilingual Speech Recognition with Corpus Relatedness Sampling (2019, background): “English TED (ted) [22] broadcast 100,000 Mandarin Hkust (hk) [23] telephone 100,000 English Switchboard (swbd)[24] telephone 100,000 Mandarin SLR18 (s18) [25] read 13,388 English Librispeech (libri) [26] read 100,000 Mandarin LDC98S73 (hub) broadcast 35,999 English Fisher (fisher) [27] telephone 100”
Mitigating Noisy Inputs for Question Answering (2019, methodology): “We then decode these utterances using the Kaldi chain model [7], trained on the Fischer-English dataset [8] for consistency with past results on mitigating ASR errors in MT [6].”
Semi-supervised maximum mutual information training of deep neural network acoustic models (2015, methodology): “In this paper, we report experiments on a subset of the Fisher English corpus [31] and several Babel languages in the LimitedLP condition.”
The IBM 2016 Speaker Recognition System (2016, methodology): “The training is accomplished using the IBM Attila toolkit [18] on 600 hours of conversational telephone speech (CTS) data from the Fisher corpus [19] with a 9-frame context of 40-dimensional speaker-adapted feature vectors obtained through per recording fMLLR transforms [10, 11].”
AISHELL-4: An Open Source Dataset for Speech Enhancement, Separation, Recognition and Speaker Diarization in Conference Scenario (2021, background): “Fisher [12] is another conversational telephone speech dataset which has yielded more than 16,000 English conversations and 2,000 hours of speech in duration, covering various speaker and speaking styles.”
Semi-Supervised Training of Acoustic Models Using Lattice-Free MMI (2018, neutral): “We report results on the Fisher English corpus [24].”