The 2016 KIT IWSLT Speech-to-Text Systems for English and German

The 2016 KIT IWSLT Speech-to-Text Systems for English and German Thai-Son Nguyen, Markus M¨uller, Matthias Sperber, Thomas Zenkel, Kevin Kilgour, Sebastian St¨uker and Alex Waibel Institute for Anthropomatics, Karlsruhe Institute of Technology Karlsruhe, Germany {first.lastname}@kit.edu Abstract This paper describes our German and English Speech- to-Text (STT) systems for the 2016 IWSLT evaluation cam- paign. The campaign focuses on the transcription of unseg- mented TED talks. Our setup includes systems using both the Janus and Kaldi frameworks. We combined the out- puts using both ROVER [1] and confusion network combina- tion (CNC) [2] to archieve a good overall performance. The individual subsystems are built by using different speaker- adaptive feature combination (e.g., lMEL with i-vector or bottleneck speaker vector), acoustic models (GMM or DNN) and speaker adaption (MLLR or fMLLR). Decoding is per- formed in two stages, where the GMM and DNN systems are adapted on the combination of the first stage outputs using MLLR, and fMLLR. The combination setup produces a final hypothesis that has a significantly lower WER than any of the individual sub- systems. For the English TED task, our best combination system has a WER of 7.8% on the development set while our other combinations gained 21.8% and 28.7% WERs for the English and German MSLT tasks. 1. Introduction For many years now, the International Workshop on Spo- ken Language Translation (IWSLT) offers a comprehensive evaluation campaign on spoken language translation. The evaluation is organized in different evaluation tracks cover- ing automatic speech recognition (ASR), machine transla- tion (MT), and the full-fledged combination of the two of them into speech translation systems (SLT). The evaluations in the tracks are conducted on TED Talks1, short 5-25min presentations by people from various fields related in some way to Technology, Entertainment, and Design (TED) [3]. In this years installment, an additional track was added using recordings from Skype (MSLT). The goal of the TED ASR track is the automatic tran- scription of fully unsegmented TED lectures. The quality of the resulting transcriptions is measured in word error rate (WER). 1http://www.ted.com/talks This system paper describes our English and German ASR setups with which we participated in the TED ASR and MSLT tracks of the 2016 IWSLT evaluation campaign. Similar to previous years’ evaluation [4], we used the Janus Recognition Toolkit (JRTk) [5] which features the IBIS single-pass decoder [6] to build several complementary sub- systems and combined them with an additional system based on the TED-LIUM recipe of the Kaldi toolkit [7]. Our Janus- based systems employ different speaker-adaptive features, acoustic models or speaker adaption techniques. While the Kaldi-based system employs the same training database, se- quence training and RNN based language models for rescor- ing. The rest of this paper is structured as follows. Section 2 describes the data that our system was trained and tested on. This is followed by Section 3 which provides a description of the acoustic front-ends used in our system and Section 7 which describes our segmentation setup. An overview of the techniques used to …

Content truncated to 500 words