Developing State-of-the-Art End-to-End ASR for Norwegian
Identifikátory výsledku
Kód výsledku v IS VaVaI
<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F46747885%3A24220%2F23%3A00012033" target="_blank" >RIV/46747885:24220/23:00012033 - isvavai.cz</a>
Výsledek na webu
<a href="https://dl.acm.org/doi/abs/10.1007/978-3-031-40498-6_18" target="_blank" >https://dl.acm.org/doi/abs/10.1007/978-3-031-40498-6_18</a>
DOI - Digital Object Identifier
<a href="http://dx.doi.org/10.1007/978-3-031-40498-6_18" target="_blank" >10.1007/978-3-031-40498-6_18</a>
Alternativní jazyky
Jazyk výsledku
angličtina
Název v původním jazyce
Developing State-of-the-Art End-to-End ASR for Norwegian
Popis výsledku v původním jazyce
We present the process of developing a modern end-to-end (E2E) automatic speech recognition (ASR) system for Norwegian (NO), which is a challenging language with many dialects and two written standards (Bokmål and Nynorsk). Since the existing speech corpora for this language are severely limited, we have had to acquire large amounts of additional data. This acquisition has been done by automatic processing of publicly accessible broadcast and parliament archives, YouTube and podcast channels, and also audiobooks. The data-harvesting process has been controlled by the ASR system, whose model has continuously been updated on the extracted chunks of speech. The final model has been trained on 1,246 h of Norwegian and further enhanced by transfer learning from an existing Swedish model. The performance of the ASR system has been evaluated on an 18-h collection of test sets (most of them publicly available) representing different application areas. Our best word error rate (WER) achieved on this collection is 7.6%, which is better than the results obtained from Google and Microsoft cloud services.
Název v anglickém jazyce
Developing State-of-the-Art End-to-End ASR for Norwegian
Popis výsledku anglicky
We present the process of developing a modern end-to-end (E2E) automatic speech recognition (ASR) system for Norwegian (NO), which is a challenging language with many dialects and two written standards (Bokmål and Nynorsk). Since the existing speech corpora for this language are severely limited, we have had to acquire large amounts of additional data. This acquisition has been done by automatic processing of publicly accessible broadcast and parliament archives, YouTube and podcast channels, and also audiobooks. The data-harvesting process has been controlled by the ASR system, whose model has continuously been updated on the extracted chunks of speech. The final model has been trained on 1,246 h of Norwegian and further enhanced by transfer learning from an existing Swedish model. The performance of the ASR system has been evaluated on an 18-h collection of test sets (most of them publicly available) representing different application areas. Our best word error rate (WER) achieved on this collection is 7.6%, which is better than the results obtained from Google and Microsoft cloud services.
Klasifikace
Druh
D - Stať ve sborníku
CEP obor
—
OECD FORD obor
10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)
Návaznosti výsledku
Projekt
<a href="/cs/project/TO01000027" target="_blank" >TO01000027: NORDTRANS - Technologie pro automatický přepis řeči ve vybraných severských jazycích</a><br>
Návaznosti
P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)
Ostatní
Rok uplatnění
2023
Kód důvěrnosti údajů
S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů
Údaje specifické pro druh výsledku
Název statě ve sborníku
Lecture Notes in Computer Science - including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics
ISBN
978-303140497-9
ISSN
03029743
e-ISSN
—
Počet stran výsledku
14
Strana od-do
200-213
Název nakladatele
Springer Science and Business
Místo vydání
—
Místo konání akce
Plzeň, ČR
Datum konání akce
1. 1. 2023
Typ akce podle státní příslušnosti
WRD - Celosvětová akce
Kód UT WoS článku
—