metadata

language:
  - gl
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - gl
  - robust-speech-event
  - model_for_talk
datasets:
  - mozilla-foundation/common_voice_8_0
model-index:
  - name: Akashpb13/Galician_xlsr
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: kmr
        metrics:
          - name: Test WER
            type: wer
            value: 0.11308483789555426
          - name: Test CER
            type: cer
            value: 0.023982371794871796
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: gl
        metrics:
          - name: Test WER
            type: wer
            value: 0.11308483789555426
          - name: Test CER
            type: cer
            value: 0.023982371794871796

Akashpb13/xlsr_hungarian_new

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_7_0 - hu dataset. It achieves the following results on the evaluation set (which is 10 percent of train data set merged with invalidated data, reported, other, and dev datasets):

Loss: 0.137096
Wer: 0.196230

Model description

"facebook/wav2vec2-xls-r-300m" was finetuned.

Intended uses & limitations

More information needed

Training and evaluation data

Training data - Common voice Galician train.tsv, dev.tsv, invalidated.tsv, reported.tsv, other.tsv and validated.tsv Only those points were considered where upvotes were greater than downvotes and duplicates were removed after concatenation of all the datasets given in common voice 7.0

Training procedure

For creating the training dataset, all possible datasets were appended and 90-10 split was used.

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 0.000096
train_batch_size: 16
eval_batch_size: 16
seed: 13
gradient_accumulation_steps: 2
lr_scheduler_type: cosine_with_restarts
lr_scheduler_warmup_steps: 500
num_epochs: 100
mixed_precision_training: Native AMP

Training results

Step	Training Loss	Validation Loss	Wer
500	5.004400	2.960605	1.000000
1000	1.653100	0.248843	0.354571
1500	0.259100	0.149203	0.251272
2000	0.155200	0.142355	0.227521
2500	0.118900	0.134033	0.217154
3000	0.100200	0.134676	0.216588
3500	0.085800	0.138649	0.219416
4000	0.075700	0.138660	0.212441
4500	0.066200	0.142651	0.208671
5000	0.060300	0.136673	0.204713
5500	0.054600	0.132755	0.202828
6000	0.048100	0.136589	0.198115
6500	0.044800	0.140990	0.199246
7000	0.039700	0.136947	0.196984
7500	0.040200	0.140098	0.196418
8000	0.037800	0.137096	0.196230

Framework versions

Transformers 4.16.0.dev0
Pytorch 1.10.0+cu102
Datasets 1.17.1.dev0
Tokenizers 0.10.3

Evaluation Commands

To evaluate on mozilla-foundation/common_voice_8_0 with split test

python eval.py --model_id Akashpb13/Galician_xlsr --dataset mozilla-foundation/common_voice_8_0 --config gl --split test