The Kaldi Speech Recognition Toolkit

By Daniel Povey, Arnab Ghoshal, Gilles Boulianne, Lukáš Burget, Ondrej Glembek, Nagendra Kumar Goel, Mirko Hannemann, Petr Motlícek, Yanmin Qian, Petr Schwarz and 3 more. 2011

6,607 citations recorded by the source corpus.

Subfields: 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 Speech & Audio

Citations in the corpus

Representation Learning with Contrastive Predictive Coding (2018, methodology): “Although the dataset does not provide labels other than the raw text, we obtained force-aligned phone sequences with the Kaldi toolkit [31] and pre-trained models on Librispeech 1 .”
Imputer: Sequence Modelling via Imputation and Dynamic Programming (2020, methodology): “We use Kaldi (Povey et al., 2011) to generate 80-dimensional filter banks, with delta and deltadelta.”
BigSSL: Exploring the Frontier of Large-Scale Semi-Supervised Learning for Automatic Speech Recognition (2021, methodology): “The inference results on the test sets are scored via corresponding Kaldi scripts, while for the Common Voice set, we take the extra step of dropping punctuation before evaluating the word error rates.”
GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10, 000 Hours of Transcribed Audio (2021, methodology): “This section describes the baseline systems and experimental results for four popular speech recognition toolkits, namely Athena, ESPnet [15], Kaldi [12] and Pika [16].”
Unsupervised Speech Recognition (2021, methodology): “Phoneme error rates are typically computed on the 39-phoneme inventory (Povey et al., 2011), which we map the phonetic transcripts to for training.”
Earnings-21: A Practical Benchmark for ASR in the Wild (2021, methodology): “If this ﬁle is excluded from evaluation, our internal Kaldi model is the most accurate.”
SpeechStew: Simply Mix All Available Speech Recognition Data to Train One Large Neural Network (2021, neutral): “We score our results with Kaldi scripts [44].”
VOXLINGUA107: A Dataset for Spoken Language Recognition (2020, methodology): “For example, the standard Kaldi [2] recipe for LRE07 2 relies on 18 LDC SLR datasets that cost $15400 to LDC non-members 3 .”
Rethinking Evaluation in ASR: Are Our Models Robust Enough? (2020, methodology): “[14] D. Povey, A. Ghoshal, G. Boulianne, L. Burget, O. Glembek, N. Goel, M. Hannemann, P. Motlicek, Y. Qian, P. Schwarz, et al., “The Kaldi speech recognition toolkit,” in ASRU, 2011.”
Fairseq S2T: Fast Speech-to-Text Modeling with Fairseq (2020, methodology): “Data Pre-Processing FAIRSEQ S2T extracts Kaldi-compliant (Povey et al., 2011) speech features (e.g. log mel-ﬁlter banks) automatically from WAV/FLAC audio ﬁles via PyKaldi (Can et al., 2018) or torchaudio 1 .”
CHiME-6 Challenge: Tackling Multispeaker Speech Recognition for Unsegmented Recordings (2020, positive): “These baselines and related implementations are integrated in the Kaldi speech recognition toolkit [36] as a recipe.”
SpecAugment: A Simple Data Augmentation Method for Automatic Speech Recognition (2019, methodology): “For Switchboard 300h, we use the Kaldi [40] “s5c” recipe to process our data, but we adapt the recipe to use 80-dimensional filter banks with delta and delta-delta acceleration.”
TED-LIUM 3: twice as much data and corpus repartition for experiments on speaker adaptation (2018, methodology): “The text from each .stm ﬁle was automatically aligned to the corresponding .sph ﬁle using the Kaldi toolkit [8].”
Librispeech: An ASR corpus based on public domain audio books (2015, background): “The corpus is freely available 4 under the very permissive CC BY 4.0 license [3] and there are example scripts in the open source Kaldi ASR toolkit [4] that demonstrate how high quality acoustic models can be trained on this data.”
Unsupervised Pre-Training of Bidirectional Speech Encoders via Masked Reconstruction (2020, methodology): “The phone-based system uses 351 position-dependent phones, generated by the Kaldi s5 recipe [25].”
Multi-Task Self-Supervised Learning for Robust Speech Recognition (2020, positive): “CHiME-5 results are based on Kaldi [37] and rely on a time-delay neural network (TDNN) [38, 39] acoustic model trained on PASE+ features.”
Unsupervised Speech Representation Learning Using WaveNet Autoencoders (2019, methodology): “The ground-truth phoneme boundaries were obtained from forced alignments using the Kaldi tri6b model from the s5 LibriSpeech recipe [55].”
Wav2Letter++: A Fast Open-source Speech Recognition System (2018, methodology): “The Kaldi Speech Recognition Toolkit [1] is by far the oldest of the aforementioned and consists of a set of standalone command-line tools.”
Light Gated Recurrent Units for Speech Recognition (2018, methodology): “The RNN part of the ASR system was implemented with Theano [63], that was coupled with the Kaldi decoder [53] to form a context-dependent RNN-HMM speech recognizer.”