Hugging Face's logo

Plim
/
xls-r-1b-cv_8-fr

Automatic Speech Recognition Transformers PyTorch
French wav2vec2 mozilla-foundation/common_voice_8_0 generated_from_trainer robust-speech-event hf-asr-leaderboard Eval Results Inference Endpoints
Model card Files Community
xls-r-1b-cv_8-fr / README.md
anton-l's picture
anton-l HF staff
Upload README.md
14d95db
preview code
|
raw history blame contribute delete
No virus
4.36 kB
metadata
language:
  - fr
license: apache-2.0
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_8_0
  - generated_from_trainer
  - robust-speech-event
  - hf-asr-leaderboard
datasets:
  - mozilla-foundation/common_voice_8_0
model-index:
  - name: XLS-R-1B - French
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 8
          type: mozilla-foundation/common_voice_8_0
          args: fr
        metrics:
          - name: Test WER (with LM)
            type: wer
            value: 15.4
          - name: Test CER (with LM)
            type: cer
            value: 5.36
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: fr
        metrics:
          - name: Test WER (with LM)
            type: wer
            value: 25.05
          - name: Test CER (with LM)
            type: cer
            value: 12.45
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Test Data
          type: speech-recognition-community-v2/eval_data
          args: fr
        metrics:
          - name: Test WER
            type: wer
            value: 27.1

Model description

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - FR dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 7.5e-05
  • train_batch_size: 16
  • eval_batch_size: 16
  • seed: 42
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 128
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 2000
  • num_epochs: 6.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.9827 0.29 1000 inf 0.2937
1.0203 0.57 2000 inf 0.2711
1.0048 0.86 3000 inf 0.2620
0.9858 1.15 4000 inf 0.2522
0.9709 1.43 5000 inf 0.2365
0.9347 1.72 6000 inf 0.2332
0.9256 2.01 7000 inf 0.2261
0.8936 2.29 8000 inf 0.2203
0.877 2.58 9000 inf 0.2096
0.8393 2.87 10000 inf 0.2017
0.8156 3.15 11000 inf 0.1936
0.8015 3.44 12000 inf 0.1880
0.774 3.73 13000 inf 0.1834
0.8372 4.01 14000 inf 0.1934
0.8075 4.3 15000 inf 0.1923
0.8069 4.59 16000 inf 0.1877
0.8064 4.87 17000 inf 0.1955
0.801 5.16 18000 inf 0.1891
0.8022 5.45 19000 inf 0.1895
0.792 5.73 20000 inf 0.1854

It achieves the best result on the validation set on STEP 13000:

  • Wer: 0.1834

Some problem occurs when calculating the validation loss.

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.3.dev0
  • Tokenizers 0.11.0

Evaluation Commands

  1. To evaluate on mozilla-foundation/common_voice_8 with split test
python eval.py --model_id Plim/xls-r-1b-cv_8-fr --dataset mozilla-foundation/common_voice_8_0 --config fr --split test
  1. To evaluate on speech-recognition-community-v2/dev_data
python eval.py --model_id Plim/xls-r-1b-cv_8-fr --dataset speech-recognition-community-v2/dev_data --config fr --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Evaluation Results

Without LM:

Dataset WER CER
TEST CV 18.33 5.60
DEV audio 31.33 13.20
TEST audio / /

With LM:

Dataset WER CER
TEST CV 15.40 5.36
DEV audio 25.05 12.45
TEST audio / /