The HTK book

By Susannah Young, J. Odell, Dg Ollason, Philip C. Woodland. 1995

How researchers see this paper

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

1. Yoshua Bengio
2. Li Deng
3. Wei-Ning Hsu
4. Dong Yu
5. Mohit Bansal
6. Jianfeng Gao
7. Quoc V. Le
8. Kyunghyun Cho

Citations in the corpus

• A3T: Alignment-Aware Acoustic and Text Pretraining for Speech Synthesis and Editing (2022, positive): “The forced alignment and G2P are both carried out by HTK (Young et al., 2002) to convert English words to phones and align phones with audio segments.”
• ESPnet: End-to-End Speech Processing Toolkit (2018, background): “Especially, these efforts have been driven by popular products including Google voice search, Amazon Alexa, and Apple Siri and open source activities including Kaldi [1], HTK [2], Sphinx [3], Julius [4], RASR [5] in addition to general research activities.”
• Estimating phoneme class conditional probabilities from raw speech signal using convolutional neural networks (2013, methodology): “They were computed (with HTK [19]) using a 25 ms Hamming window on the speech signal, with a shift of 10 ms.”
• Analysis of CNN-based speech recognition system using raw speech as input (2015, methodology): “The alignment is obtained using the HTK-based HMM/GMM system provided along with the database.”
• An end-to-end generative framework for video segmentation and recognition (2015, background): “Structured temporal models, on the other hand, have reached an impressive level of maturity in several engineering domains and speech recognition [36] in particular.”
• Convolutional Neural Networks-based continuous speech recognition using raw speech signal (2015, methodology): “They were computed (with HTK [23]) using a 25 ms Hamming window on the speech signal, with a shift of 10 ms.”
• Speech and Language Processing: An Introduction to Natural Language Processing, Computational Linguistics, and Speech Recognition (2000, background): “Language models for LVCSR tend to be trigrams or even fourgrams; good toolkits are available to build and manipulate them (Stolcke, 2002; Young et al., 2005).”
• Model architectures to extrapolate emotional expressions in DNN-based text-to-speech (2021, methodology): “The phoneme duration was manually annotated and labeled with the same format as the hidden Markov model toolkit (HTK) [24].”
• Hybrid Deep Neural Network--Hidden Markov Model (DNN-HMM) Based Speech Emotion Recognition (2013, methodology): “For each of the emotions in the database, the HTK toolkit [15] has been used to train a baseline left-right GMMHMM with 5 states and 17 Gaussian mixtures.”
• Deep Belief Networks using discriminative features for phone recognition (2011, methodology): “For Gaussian-binary RBMs, we ran 150 epochs with a fixed learning rate of 0.005 while for binary-binary RBMs we used 50 epochs with a learning rate of 0.08.”
• End-to-end Phoneme Sequence Recognition using Convolutional Neural Networks (2013, methodology): “They were computed (with HTK [28]) using a 25 ms Hamming window on the speech signal, with a shift of 10 ms.”
• Wespeaker: A Research and Production Oriented Speaker Embedding Learning Toolkit (2022, background): “The early general speech processing toolkits, such as HTK [5] and Kaldi [6], had equipped many researchers and industrial productions before deep learning toolkits such as Pytorch [7] and Tensorflow [8], while the recently thriving Pytorch based SpeechBrain [9], Espnet [10], are much more friendly…”
• The Voice Conversion Challenge 2018: Promoting Development of Parallel and Nonparallel Methods (2018, methodology): “WERs were calculated by the HResults tool in HTK [42] using manual transcriptions as ground truth.”
• A deep learning approach for generalized speech animation (2017, methodology): “For speech-to-text transcription, we used either off-theshelf software such as the Penn Phonetics Lab Forced Aligner [Yuan and Liberman 2008] that is based on the HTK toolbox [Young et al. 2009], or manual transcription in special cases.”
• Building DNN acoustic models for large vocabulary speech recognition (2014, methodology): “A full overview of LVCSR systems is beyond the scope of this work, so we instead refer to previous articles for an overview of HMM-based speech recognition systems [8], [9], [10], [11], [12].”
• Speaker verification and spoken language identification using a generalized i-vector framework with phonetic tokenizations and tandem features (2014, methodology): “2, HTK toolkit [23] is used to decode the PPP features and output a lattice file for each utterance which is further processed into n-gram counts and n-gram indexes by the lattice-tool toolkit [24].”