metadata
language:
- sv-SE
license: cc0-1.0
tags:
- automatic-speech-recognition
- mozilla-foundation/common_voice_8_0
- generated_from_trainer
- sv
- robust-speech-event
- model_for_talk
datasets:
- mozilla-foundation/common_voice_8_0
- marinone94/nst_sv
model-index:
- name: XLS-R-300M - Swedish
results:
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: mozilla-foundation/common_voice_8_0
type: mozilla-foundation/common_voice_8_0
args: sv-SE
metrics:
- name: Test WER
type: wer
value: 16.98
- name: Test CER
type: cer
value: 5.66
- task:
name: Automatic Speech Recognition
type: automatic-speech-recognition
dataset:
name: speech-recognition-community-v2/dev_data
type: speech-recognition-community-v2/dev_data
args: sv
metrics:
- name: Test WER
type: wer
value: 27.01
- name: Test CER
type: cer
value: 13.14
This model is a fine-tuned version of KBLab/wav2vec2-large-voxrex on 2 epochs of the MARINONE94/NST_SV - SV dataset (80% random split with seed 42 as the dataset for now has only the "train" split), and then on 50 epochs of the the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - SV-SE dataset ("train+validation" split). See run.sh to have a complete overview of all the training steps. NOTE: the first training for now didn't work as expected, so it might be useless or even degrade performance. Further investigation and development is needed.
d73da225cfdc57213ea4ab67b24bb87ac41f4392 is the commit at the end of the first training:
sh run.sh