metadata

language:
  - ha
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - ha
  - robust-speech-event
  - model_for_talk
datasets:
  - mozilla-foundation/common_voice_8_0
model-index:
  - name: Akashpb13/Hausa_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: 0.21177149451817656
          - name: Test CER
            type: cer
            value: 0.04472733528298641
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: kmr
        metrics:
          - name: Test WER
            type: wer
            value: 0.21177149451817656
          - name: Test CER
            type: cer
            value: 0.04472733528298641

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: 0.275118
Wer: 0.329955

Model description

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

Intended uses & limitations

More information needed

Training and evaluation data

Training data - Common voice Hausa train.tsv, dev.tsv, invalidated.tsv, reported.tsv and other.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: 50
mixed_precision_training: Native AMP

Training results

Step	Training Loss	Validation Loss	Wer
500	5.176500	2.776366	1.000000
1000	1.166500	0.329668	0.482262
1500	0.334500	0.245564	0.391304
2000	0.226300	0.241998	0.381168
2500	0.174600	0.227778	0.356895
3000	0.141700	0.237551	0.343292
3500	0.117600	0.259884	0.348626
4000	0.097800	0.261468	0.343825
4500	0.083500	0.268763	0.332889
5000	0.074500	0.276381	0.332622
5500	0.067700	0.275504	0.328354
6000	0.064500	0.274393	0.331288
6500	0.061400	0.275118	0.329955

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/Hausa_xlsr --dataset mozilla-foundation/common_voice_8_0 --config ha --split test