Phone Recognition on the TIMIT Database

By Carla Lopes, Fernando Perdigão. 2012

How researchers see this paper

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

1. Yonatan Belinkov
2. Yoshua Bengio
3. Xu Tan
4. Mohit Bansal
5. Wei-Ning Hsu
6. Sergey Levine
7. Dong Yu
8. Kyunghyun Cho

Citations in the corpus

• Beyond Sparsity: Tree Regularization of Deep Models for Interpretability (2017, background): “• Phonetic Speech (TIMIT): We have recordings of 630 speakers of eight major dialects of American English reading ten phonetically rich sentences (Garofolo et al. 1993).”
• Cascaded Cross-Module Residual Learning towards Lightweight End-to-End Speech Coding (2019, methodology): “300 and 50 speakers are randomly selected from TIMIT [32] training and test datasets, respectively.”
• Streaming Sparse Gaussian Process Approximations (2017, methodology): “The second dataset is an audio signal prediction dataset, produced from the TIMIT database [23] and previously used to evaluate GP approximations [24].”
• Composition of Deep and Spiking Neural Networks for Very Low Bit Rate Speech Coding (2016, background): “Segmental features — phonological posteriors — have values mostly concentrated very close to either 1 or 0, and these binary patterns allow very efficient use of 1-bit quantization: the probabilities above 0.5 are normalized to 1 and the probabilities less than 0.5 are forced to zero.”
• CVSS Corpus and Massively Multilingual Speech-to-Speech Translation (2022, background): “…a mixture of a proprietary multilingual speech query dataset covering 37 locales collected by vendors, plus public corpora including LibriVox, CN-Celeb (Fan et al., 2020), TIMIT (Garofolo et al., 1993), Fisher (Cieri et al., 2004), and Mixer 4 and 5 (Cieri et al., 2007; Brandschain et al., 2008).”
• An End-to-End Multimodal Voice Activity Detection Using WaveNet Encoder and Residual Networks (2019, background): “and speaker recognition [12] and even multimodal tasks such as visual question answering [13].”
• Joint Phoneme Alignment and Text-Informed Speech Separation on Highly Corrupted Speech (2020, methodology): “We use the instrumental accompaniments of the MUSDB18 data set [22] as music signals and mix them with the TIMIT [23] speech signals.”
• Asynchronous stochastic gradient descent for DNN training (2013, methodology): “For more details about TIMIT, we refer the readers to [7].”
• Multidimensional Residual Learning Based on Recurrent Neural Networks for Acoustic Modeling (2016, methodology): “In the endto-end system trained with CTC criterion, the results of unidirectional model and bidirectional model on TIMIT task are cited from [20].”
• Survey on Deep Neural Networks in Speech and Vision Systems (2019, background): “TIMT includes 630 speakers from several American English dialects [147].”
• Learning Robust and Multilingual Speech Representations (2020, methodology): “In addition we used the audio from TIMIT (Garofolo, 1993) and the Speech Accent Archive (Weinberger and Kunath, 2009), ignoring the transcrip- 1https://nvidia.github.io/OpenSeq2Seq/ html/speech-recognition/wave2letter.html 2https://voice.mozilla.org tions.”
• An open-source end-to-end ASR system for Brazilian Portuguese using DNNs built from newly assembled corpora (2020, background): “For the sake of comparison, the most important datasets employed in the ASR literature are the 5.4-h TIMIT [45], the 73-h Wall Street Journal (WSJ) [46], [47], the 300-h Switchboard [48], and the 1, 000-h Librispeech [39], all in English.”
• Can Speaker Augmentation Improve Multi-Speaker End-to-End TTS? (2020, background): “TIMIT [33]: Speakers of eight American English dialects each read ten phonetically-rich sentences.”
• Speech Dereverberation Using Non-Negative Convolutive Transfer Function and Spectro-Temporal Modeling (2016, methodology): “As clean speech signals, 16 different sentences (uttered by 16 speakers) from the TIMIT database [34] were used to make the results independent of the speech material.”
• A Brief Overview of Unsupervised Neural Speech Representation Learning (2022, background): “1 seconds on average (Garofolo 1993), whereas spectrogram frames xt are typically computed at 0.”
• An auto-encoder based approach to unsupervised learning of subword units (2014, methodology): “We used 3 datasets, one extracted from the Italian Lecce dataset [16] and 2 from the TIMIT corpus [17].”