Baybars's picture
evaluation results
52ce928
|
raw
history blame
3.19 kB
metadata
language:
  - tr
license: apache-2.0
tags:
  - automatic-speech-recognition
  - common_voice
  - generated_from_trainer
  - tr
  - robust-speech-event
datasets:
  - common_voice
model-index:
  - name: ''
    results: []

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

  • Loss: 0.4164
  • Wer: 0.3098
  • Cer: 0.0764

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Language Model

N-gram language model is trained by mpoyraz on a Turkish Wikipedia articles using KenLM and ngram-lm-wiki repo was used to generate arpa LM and convert it into binary format.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

  • learning_rate: 0.0005
  • train_batch_size: 64
  • eval_batch_size: 8
  • seed: 42
  • optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • num_epochs: 100.0
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer Cer
0.6356 9.09 500 0.5055 0.5536 0.1381
0.3847 18.18 1000 0.4002 0.4247 0.1065
0.3377 27.27 1500 0.4193 0.4167 0.1078
0.2175 36.36 2000 0.4351 0.3861 0.0974
0.2074 45.45 2500 0.3962 0.3622 0.0916
0.159 54.55 3000 0.4062 0.3526 0.0888
0.1882 63.64 3500 0.3991 0.3445 0.0850
0.1766 72.73 4000 0.4214 0.3396 0.0847
0.116 81.82 4500 0.4182 0.3265 0.0812
0.0718 90.91 5000 0.4259 0.3191 0.0781
0.019 100.0 5500 0.4164 0.3098 0.0764

Evaluation Commands

Please install unicode_tr package before running evaluation. It is used for Turkish text processing.

  1. To evaluate on mozilla-foundation/common_voice_7_0 with split test
python eval.py --model_id Baybars/wav2vec2-xls-r-300m-cv8-turkish --dataset mozilla-foundation/common_voice_8_0 --config tr --split test
  1. To evaluate on speech-recognition-community-v2/dev_data
python eval.py --model_id Baybars/wav2vec2-xls-r-300m-cv8-turkish --dataset speech-recognition-community-v2/dev_data --config tr --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Framework versions

  • Transformers 4.17.0.dev0
  • Pytorch 1.10.2+cu102
  • Datasets 1.18.2.dev0
  • Tokenizers 0.11.0