Interspeech 2009 2009
DOI: 10.21437/interspeech.2009-604
|View full text |Cite
|
Sign up to set email alerts
|

The RWTH aachen university open source speech recognition system

Abstract: We announce the public availability of the RWTH Aachen University speech recognition toolkit. The toolkit includes state of the art speech recognition technology for acoustic model training and decoding. Speaker adaptation, speaker adaptive training, unsupervised training, a finite state automata library, and an efficient tree search decoder are notable components. Comprehensive documentation, example setups for training and recognition, and a tutorial are provided to support newcomers.

Help me understand this report

Search citation statements

Order By: Relevance

Paper Sections

Select...
1
1
1
1

Citation Types

0
5
0

Year Published

2011
2011
2014
2014

Publication Types

Select...
5
3

Relationship

0
8

Authors

Journals

citations
Cited by 56 publications
(5 citation statements)
references
References 25 publications
0
5
0
Order By: Relevance
“…In order to achieve this goal, we decided to keep the amount of parameter optimization as low as possible. This system is based on the software developed at RWTH Aachen University: RASR (Rybach et al 2011;Wiesler et al 2014) and RETURNN (Doetsch et al 2017;Zeyer, Alkhouli, and Ney 2018).…”
Section: Open-condition Systemmentioning
confidence: 99%
“…In order to achieve this goal, we decided to keep the amount of parameter optimization as low as possible. This system is based on the software developed at RWTH Aachen University: RASR (Rybach et al 2011;Wiesler et al 2014) and RETURNN (Doetsch et al 2017;Zeyer, Alkhouli, and Ney 2018).…”
Section: Open-condition Systemmentioning
confidence: 99%
“…They have been used recently for describing deep learning models [Abadi et al 2015]. We can also find several systems influenced by this dataflow architecture in the speech recognition field: for instance, the ATK wrapper for HTK [Young 2007] or also the preprocessing step of the RWTH speech recognition system 4 [Rybach et al 2009].…”
Section: Richard Feynmanmentioning
confidence: 99%
“…By taking into account the best LM transition, in the case of LM pruning, or the best HMM transition at sub-word level together with an estimate of the best emission probabilities (computed from the training corpus and kept constant [Sixtus 2003; Section 9.1], or computed from the current frame using cheaper models as in the fast-match lookahead technique described in Section 10.6.2). In the case of [Rybach 2014;Section 6.3.1], the pruning is applied before computing the acoustic model scores and the hypothesis is compared with a current threshold which does not contain the acoustic score either in order to be more properly compared.…”
Section: Beam Searchmentioning
confidence: 99%
See 1 more Smart Citation
“…In the following we will demonstrate on a Large Vocabulary Continuous Speech Recognition (LVCSR) task, how these signal distortions influence the ASR performance. Experiment was made using the RWTH Aachen University open source ASR system [26], and Catalan Speecon and FreeSpeech databases. For training, approximately 121 hours of data from both databases were selected, using only close-talking Speech signal was framed applying 25 ms long Hamming window with 10 ms overlap.…”
Section: The Effects Of Reverberation On the Asr Performancementioning
confidence: 99%