SLT.KIT

This repository contains a Spoken Language Translation System. It can be used to translate the output of an Automatic Speech Recognition (ASR) system. The system contains of an monolingual translation system that adds punctuation marks to the output of the ASR system. Furthermore, it recases the output. Then the output is translated by an machine translation system system. The system can be used to train such system as well as pre-trained systems are availabel. The systems can be trained and used by installing the docker container.

The system uses the following software:

• OpenNMT-py
• Moses
• Subword NMT
• Translation error rate
• BEER
• CharacTER
• NLTK

Requirements:

• Docker

Installation

	git clone https://github.com/jniehues-kit/SLT.KIT.git
	cd SLT.KIT
	docker build -t slt.kit -f Dockerfile.ST-Baseline .

Run

• Starting the docker container (e.g. source language English (en) and target language German (de))

	docker run -ti --rm --runtime=nvidia -e NVIDIA_VISIBLE_DEVICES=$gpuid slt.kit
	export sl=en
	export tl=de

• Within a docker container

  1 Training a model or download pre-trained model

  • Training a model

	/opt/SLT.KIT/systems/${model}/Train.sh

* Download a pre-trained model

	/opt/SLT.KIT/systems/${model}/Download.sh

2 Translate test set

	/opt/SLT.KIT/systems/${model}/Translate.sh $testset

where $model can be:

• smallTED

and $testset can be:

• English to {German | French} ** dev2010 ** tst2010 ** tst2013 (only German) ** tst2014 ** tst2015
• German to English ** dev2012 ** tst2013 ** tst2014

Models

small TED

SLT model trained only on the [TED corpus] (https://wit3.fbk.eu/)

Peformance:

English to German

SET	BLEU	TER	BEER	CharacTER	BLEU(ci)	TER(ci)
dev2010	14.46	70.98	46.61	83.77	15.42	69.00
dev2010 (manual Transcript)	23.45	56.74	54.44	56.77	25.03	55.17
tst2010	10.41	76.53	36.15	318.59	11.04	74.96
tst2010 (manual Transcript)	24.81	55.66	53.34	55.85	26.41	54.04
tst2013	13.91	71.71	44.54	80.07	14.81	69.60
tst2013 (manual Transcript)	26.05	54.27	54.34	54.22	27.49	52.98
tst2014	13.24	72.34	43.78	83.44	14.03	70.57
tst2014 (manual Transcript)	22.31	58.36	51.85	57.66	23.18	57.44
tst2015	13.03	83.20	43.66	74.03	13.75	81.30
tst2015 (manual Transcript)	25.07	57.76	53.10	54.77	26.06	56.81

English to French

SET	BLEU	TER	BEER	CharacTER	BLEU(ci)	TER(ci)
dev2010	13.6	78.37	45.80	74.62	14.37	76.65
dev2010 (manual Transcript)	22.85	64.49	53.09	56.77	24.04	63.22
tst2014	16.99	69.48	47.42	77.37	17.88	67.77
tst2014 (manual Transcript)	28.64	54.90	56.15	52.08	29.36	54.16
tst2015	17.17	71.06	47.30	70.50	18.15	69.18
tst2015 (manual Transcript)	28.46	56.21	56.21	50.68	29.26	55.28

German to English

SET	BLEU	TER	BEER	CharacTER	BLEU(ci)	TER(ci)
dev2012	10.66	79.40	42.97	82.97	11.49	77.57
dev2012 (manual Transcript)	19.58	66.13	52.20	66.61	20.12	65.28
tst2013	10.76	76.36	40.52	138.57	11.53	74.67
tst2013 (manual Transcript)	22.68	59.64	56.15	57.47	23.45	58.63
tst2014	10.05	79.03	41.04	93.92	10.69	77.46
tst2014 (manual Transcript)	17.86	65.73	51.52	64.09	18.3	64.99