This dataset contains transcriptions of speech files derived from the Norwegian language corpus provided by Språkbanken, specifically the nb_samtale subset. These transcriptions have been subjected to controlled noise addition to simulate various acoustic environments.

URI: https://huggingface.co/datasets/Sprakbanken/nb_samtale/viewer/annotations/train?f[duration][min]=24.6432&f[duration][imax]=27.368

Original Audio properties:

Duration : 24 secounds - 27 secounds

Format: WAV

Number of files: 9

Transcriptions:

Number of Models tested: 3

Number of noise tested: 4

Number of noise levels: 16

TOTAL Number of files: 1682

Metrics Word Error Rate (WER): WER calculations are based on the transcriptions with 0% added noise as the ground truth. This metric helps in assessing the performance of speech recognition systems under varying noise conditions.