Publications

We have made a Norwegian, 5000+ hour dataset for weak supervision ASR (e.g. Whisper) by aligning audio from the Norwegian parliament with the official parliamentary proceedings

Per Erik Solberg, Pierre Beauguitte, Per Egil Kummervold, Freddy Wetjen

https://www.researchgate.net/publication/370766648_A_Large_Norwegian_Dataset_for_Weak_Supervision_ASR (2023)

We present a study on optimizing the variability of training material for ASR when the linguistic resources are limited.

Per Erik Solberg, Pablo Ortiz, Phoebe Parsons, Torbjørn Svendsen, Giampiero Salvi

https://aclanthology.org/2023.nodalida-1.51/ (2023)

We propose an analysis of the impact of dialect on ASR results for Norwegian.

Phoebe Parsons, Knut Kvale, Torbjørn Svendsen, Giampiero Salvi

https://aclanthology.org/2023.nodalida-1.47/ (2023)

We propose a new semantically meaninful metric based on BERT and dynamic programming for ASR evaluation.

Janine Rugayan, Torbjørn Svendsen, Giampiero Salvi

10.21437/Interspeech.2022-817 (2022)

We propose new, data-efficient training tasks for BERT models that improve performance of automatic speech recognition (ASR) systems on conversational speech.

Pablo Ortiz, Simen Burud

arXiv:2110.02267 (2021)

We investigate the problem of speaker independent acoustic-to-articulatory inversion (AAI) in noisy conditions within the deep neural network (DNN) framework.

Abdolreza Sabzishahrebabaki, Giampiero Salvi, Torbjørn Karl Svendsen, Sabato Marco Siniscalchi

10.1109/TASLP.2021.3133218

The Norwegian Parliamentary Speech Corpus (NPSC) is the first, publicly available dataset containing unscripted, Norwegian speech designed for training of automatic speech recognition (ASR) systems. The NPSC-trained system performed significantly better, with a 22.9% relative improvement in word error rate (WER) and a “democratizing” effect in terms of dialects.

Per Erik Solberg, Pablo Ortiz

Proceedings of LREC2022