python >= 3.0
- Data processing scripts from
moses
The translation, segmentation and truecasing models required by the system are available for download via Zenodo. You must download the models separately, and extract the archive into the models
folder.
translation models:
(en
=English, de
=German, a
=audio, t
=text)
models/translate_a-t.pt
BLEU: 14.38
src_tgt pairs:
en_a-de_t
en_a-en_t
models/translate_at-t.pt
BLEU: 11.71
src_tgt pairs:
en_a-de_t
en_t-de_t
en_a-en_t
de_t-en_t
translate_a-t.pt
can only translate audio to text. translate_at-t.pt
can translate both audio and text to text.
translation_example.sh
script shows examples of translating audio to text and text to text using example data.
When translating text to text, the input text has to be preprocessed with normalization, truecasing and segmentation scripts:
scripts/preprocess-01-normalize.sh LANG_ID < INPUT > OUTPUT
scripts/preprocess-02-truecase.sh TRUECASER_MODEL < INPUT > OUTPUT
scripts/preprocess-03-segment.sh SEGMENTATION_MODEL INPUT OUTPUT
The preprocessing scripts use the following parameters:
LANG_ID: en or de
TRUECASER_MODEL: models/truecaser-{en,de}.model
SEGMENTATION_MODEL: models/sentencepiece-{en,de}-bpe-32K.model
The translation output has to be postprocessed to produce plain text result:
scripts/postprocess.sh LANG_ID < INPUT > OUTPUT
Where:
LANG_ID: en or de
- The University of Helsinki Submission to the IWSLT2020 Offline Speech Translation Task
@inproceedings{vazquez-etal-2020-university,
title = "The {U}niversity of {H}elsinki Submission to the {IWSLT}2020 Offline {S}peech{T}ranslation Task",
author = {V{\'a}zquez, Ra{\'u}l and
Aulamo, Mikko and
Sulubacak, Umut and
Tiedemann, J{\"o}rg},
booktitle = "Proceedings of the 17th International Conference on Spoken Language Translation",
month = jul,
year = "2020",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.iwslt-1.10",
doi = "10.18653/v1/2020.iwslt-1.10",
pages = "95--102",
}