TIMIT Acoustic-Phonetic Continuous Speech Corpus

By Carla Lopes, Fernando Perdigão. 2012

1,364 citations recorded by the source corpus.

Subfields: 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 Evaluation & Benchmarks

In Speech & Audio

Citations in the corpus

Benchmarking Deep Reinforcement Learning for Continuous Control (2016, neutral): “…(Fei-Fei et al., 2006), CI-FAR (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 & Pearce, 2000), and VoiceSearch (Yu et al., 2007).”
Learning Robust and Multilingual Speech Representations (2020, positive): “In addition we used the audio from TIMIT (Garofolo, 1993) and the Speech Accent Archive (Weinberger and Kunath, 2009), ignoring the transcrip-”
Learning Filterbanks from Raw Speech for Phone Recognition (2017, methodology): “We perform phone recognition experiments on TIMIT [17] using the standard train/dev/test split.”
RNADE: The real-valued neural autoregressive density-estimator (2013, methodology): “We also measured the ability of RNADE to model small patches of speech spectrograms, extracted from the TIMIT dataset [29].”
A Baseline for Detecting Misclassified and Out-of-Distribution Examples in Neural Networks (2016, methodology): “Our sequence prediction system uses a bidirectional LSTM with two-layers and a clipped GELU nonlinearity, optimized for 60 epochs with RMSProp trained on 80% of the TIMIT corpus (Garofolo et al., 1993).”
Gender and Dialect Bias in YouTube’s Automatic Captions (2017, background): “TIMIT, the most-distributed corpora available through the linguistic data consortium, is balanced for speaker dialect but approximately 69% of the speech in it comes from male speakers (Garofolo et al., 1993).”
Adversarial Audio Synthesis (2018, methodology): “4 hours): Multiple speakers, clean (Garofolo et al., 1993) We train our networks using batches of size 64 on a single NVIDIA P100 GPU.”
Disentangled Sequential Autoencoder (2018, methodology): “The TIMIT data (Garofolo et al., 1993) contains broadband 16kHz recordings of phonetically-balanced read speech.”
Looking to listen at the cocktail party (2018, neutral): “For example, the CUAVE dataset [Patterson et al. 2002] Garofolo et al. 1992], and are recorded using both front-facing and 30-degree cameras.”
End-To-End Source Separation With Adaptive Front-Ends (2017, methodology): “For training the neural networks, we randomly selected 10 malefemale speaker pairs from the TIMIT database [18].”
DeepMMSE: A Deep Learning Approach to MMSE-Based Noise Power Spectral Density Estimation (2020, methodology): “…speech recordings from the following speech corpora are included in the training set: the train-clean-100 set from the Librispeech corpus [39] ( 28 539 utterances), the CSTR VCTK corpus [40] ( 42 015 utterances), and the si ∗ and sx ∗ training sets from the TIMIT corpus [41] ( 3 696 utterances).”
MEAD: A Large-Scale Audio-Visual Dataset for Emotional Talking-Face Generation (2020, methodology): “These corpora are much less diverse than the TIMIT [16] corpus used by SAVEE [23] and our dataset.”
Zero-Cost Proxies for Lightweight NAS (2021, methodology): “• NAS-Bench-ASR: This is our in-house dataset for convolution-based automatic speech recognition models evaluated on the TIMIT dataset (Garofolo et al., 1993).”
Acoustic Modeling With Hierarchical Reservoirs (2013, neutral): “The TIMIT corpus [52] contains recordings of 630 native American speakers–438 males and 192 females.”
A Task-Optimized Neural Network Replicates Human Auditory Behavior, Predicts Brain Responses, and Reveals a Cortical Processing Hierarchy. (2018, positive): “Speech and music training examples were drawn from large, labeled corpora (Bertin-Mahieux et al., 2011; Garofolo and Consortium, 1993; Paul and Baker, 1992) and were superimposed on different types of real-world background ‘‘noise’’ to make the task more challenging and realistic (see STAR Methods…”
Self-Supervised Speech Representation Learning: A Review (2022, background): “Several datasets, including GigaSpeech [220], TED-LIUM 3 (TED3) [221], TED-LIUM 2 (TED2) [222], Switchboard (SWB) [223], TIMIT [96], and VoxLingua107 [224], are labeled and conventionally used for evaluation, while their audio streams are also aggregated to build diversified and large-scale corpora ”
Phonetic posteriorgrams for many-to-one voice conversion without parallel data training (2016, methodology): “In the PPGs-based approach, the SI-ASR system is implemented using the Kaldi speech recognition toolkit [17] with the TIMIT corpus [18].”