TTS, Language Technology, Kvistur, and Samrómur papers published but no conferences to attend

Sjá íslenska þýðingu neðar

This is a positive but somewhat sad week for LVL. Many LVL members were going to go to Marseille, France this week to attend Language Resources & Evaluation Conference (LREC) 2020 and the joint Spoken Language Technologies for Under-Resourced Languages and Collaboration and Computing for Under-Resourced Languages (SLTU-CCURL) 2020 Workshop. Once there they were going to present their many papers, providing an in-depth look into our TTS, data collection, compound splitting, and general language technology research in recent months. However, due to COVID-19 these conferences were both cancelled. Luckily the organizers have still decided to publish the proceedings this month. The joint SLTU proceedings were published May 8th on the SLTU-CCURL 2020 website at Workshop Proceedings (our paper is on page 316). Head over to the SLTU 2020 website if you want to read more SLTU-CCURL papers. We’re still waiting for the LREC proceedings to be published. But our papers can now be found as pdfs below and on our publications page.

Our TTS paper was accepted at SLTU-CCURL 2020:

Title: Manual Speech Synthesis Data Acquisition – From Script Design to Recording Speech
Authors: Atli Þor Sigurgeirsson, Gunnar Thor Örnólfsson, Jon Gudnason
Summary: In this paper we present the work of collecting a large amount of high quality speech synthesis data for Icelandic. A script design strategy is proposed and three scripts have been generated to maximize diphone coverage, varying in length. The largest reading script contains 14,400 prompts and includes 81% of all Icelandic diphones at least twenty times. As of writing, 58.7 hours of high quality speech data has been collected.

Our Samrómur, Kvistur, and Language Technology programme papers were accepted at LREC 2020:

The cover of the proposal sent to the Icelandic parliament

Title: Language Technology Programme for Icelandic 2019-2023
Authors: Anna Nikulásdóttir, Jón Guðnason, Anton Karl Ingason, Hrafn Loftsson, Eiríkur Rögnvaldsson, Einar Freyr Sigurðsson and Steinþór Steingrímsson
Summary: In this paper, we describe a national language technology programme for Icelandic. The programme aims at making Icelandic usable in communication and interactions in the digital world, by developing accessible, opensource language resources and software. The research and development work within the programme is carried out by SÍM, a consortium of universities, institutions, and private companies, with a strong emphasis on cooperation between academia and industries. Five core projects will be the main content of the programme: language resources, speech recognition, speech synthesis, machine translation, and
spell and grammar checking.

A representation of the model with one BiLSTM layer, showing where the compound word raforku ‘electric energy’ is split in two.

Title: Kvistur: a BiLSTM Compound Splitter for Icelandic
Authors: Jón Daðason, David Mollberg and Hrafn Loftsson
Summary: In this paper, we present a character-based BiLSTM model for splitting Icelandic compound words, and show how quantity of training data affects model performance. Compounding is highly productive in Icelandic, and new compounds are constantly being created. This results in a large number of out-of-vocabulary (OOV) words, negatively impacting the performance of many NLP tools. Our model is trained on a dataset of 2.9 million unique word forms and their constituent structures from the Database of Icelandic Morphology. The model learns to split compound words into two and can be used to derive a word form’s constituent structure. Knowing the constituent structure of a word form makes it possible to generate the optimal split for a given task. The model outperforms other previously published methods when evaluated on a corpus of manually split word forms. This method has been integrated into Kvistur, an Icelandic compound word analyzer.

The cumulative count of votes and utterances. Each utterance can have more than one vote as it needs two positive votes to be considered valid and two negative votes to be considered invalid.

Title: Samrómur: Crowd-sourcing Data Collection for Icelandic Speech Recognition
Authors: David Erik Mollberg, Ólafur Helgi Jónsson, Sunneva Þorsteinsdóttir, Steinþór Steingrímsson, Eydís Huld Magnúsdóttir and Jon Gudnason
Summary: This contribution describes an ongoing speech data collection, using Samrómur which is built upon Mozilla’s Common Voice. The goal is to build a large-scale speech corpus for Automatic Speech Recognition (ASR) for Icelandic. Upon completion, Samrómur will be the largest open speech corpus for Icelandic. The paper discusses the methods used for crowd-sourcing and illustrate the importance of marketing and good media coverage for a crowd-sourced dataset. Preliminary results exceed our
expectations. The paper also reports on the process of validating recordings.

Our SÍM colleagues also had two papers at LREC 2020: “Facilitating Corpus Usage: Making Icelandic Corpora More Accessible for Researchers and Language Users” and “Parallel Universal Dependencies”. Congratulations!

While it is sad that our LVL members cannot meet with fellow researchers and visit the great city of Marseille, they still look forward to connecting with researchers online through your comments on their papers and links to your related papers.

Þessa vikuna hefði átt að halda Language Resources & Evaluation (LREC) ráðstefnuna í Frakklandi, sem og Spoken Language Technologies for Under-resourced Languages vinnustofuna en báðum þessum viðburðum var aflýst vegna COVID-19.  Margir starfsmenn LVL ætluðu sér að sækja þessa viðburði og kynna þar 4 greinar og veita innsýn í þær máltæknirannsóknir sem hafa farið fram hérna síðustu mánuði. Hérna má lesa nánar um þetta og nálgast greinarnar. (Athugið að greinarnar eru aðeins aðgengilegar á ensku).

Leave a Reply

%d bloggers like this: