JHU Kaldi system for Arabic MGB-3 ASR challenge using diarization, audio-transcript alignment and transfer learning

53 citations recorded by the source corpus.

How researchers see this paper

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

GigaSpeech: An Evolving, Multi-domain ASR Corpus with 10, 000 Hours of Transcribed Audio (2021, methodology): “Our aligner is implemented with Kaldi [11], and the alignment procedure follows the work here [12], which adopts a divide and conquer strategy to tackle the alignment problem.”
Rethinking Evaluation in ASR: Are Our Models Robust Enough? (2020, neutral): “In the context of the Arabic MGB-3 challenge, the authors of [2] transferred acoustic models trained on broadcast TV to Youtube videos, with a different setting than here as the training transcriptions were noisily labeled.”
Lattice-free State-level Minimum Bayes Risk Training of Acoustic Models (2018, background): “In previous literature [8, 11, 12], sMBR training after LF-MMI training showed a relative improvement of about 2%-6% 1 over original LF-MMI AMs.”
Semi-Supervised Training of Acoustic Models Using Lattice-Free MMI (2018, methodology): “The phone LM used for creating the denominator FST is estimated using phone sequences from both supervised and unsupervised data as in [28, 29].”
Investigation of transfer learning for ASR using LF-MMI trained neural networks (2017, background): “This is particularly useful in cases where the target corpus is very small compared to the source corpus, as found in the case of MGB-3 challenge [30].”