metadata

license: apache-2.0
language: tr
tags:
  - automatic-speech-recognition
  - common_voice
  - hf-asr-leaderboard
  - robust-speech-event
  - tr
datasets:
  - common_voice
model-index:
  - name: mpoyraz/wav2vec2-xls-r-300m-cv6-turkish
    results:
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Common Voice 6.1
          type: common_voice
          args: tr
        metrics:
          - name: Test WER
            type: wer
            value: 8.83
          - name: Test CER
            type: cer
            value: 2.37
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Dev Data
          type: speech-recognition-community-v2/dev_data
          args: tr
        metrics:
          - name: Test WER
            type: wer
            value: 32.81
          - name: Test CER
            type: cer
            value: 11.22
      - task:
          name: Automatic Speech Recognition
          type: automatic-speech-recognition
        dataset:
          name: Robust Speech Event - Test Data
          type: speech-recognition-community-v2/eval_data
          args: tr
        metrics:
          - name: Test WER
            type: wer
            value: 34.86

wav2vec2-xls-r-300m-cv6-turkish

Model description

This ASR model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on Turkish language.

Training and evaluation data

The following datasets were used for finetuning:

Common Voice 6.1 TR All validated split except test split was used for training.
MediaSpeech

Training procedure

To support both of the datasets above, custom pre-processing and loading steps was performed and wav2vec2-turkish repo was used for that purpose.

Training hyperparameters

The following hypermaters were used for finetuning:

learning_rate 2e-4
num_train_epochs 10
warmup_steps 500
freeze_feature_extractor
mask_time_prob 0.1
mask_feature_prob 0.1
feat_proj_dropout 0.05
attention_dropout 0.05
final_dropout 0.1
activation_dropout 0.05
per_device_train_batch_size 8
per_device_eval_batch_size 8
gradient_accumulation_steps 8

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.10.1
Datasets 1.18.3
Tokenizers 0.10.3

Language Model

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

Evaluation Commands

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

To evaluate on common_voice with split test

python eval.py --model_id mpoyraz/wav2vec2-xls-r-300m-cv6-turkish --dataset common_voice --config tr --split test

To evaluate on speech-recognition-community-v2/dev_data

python eval.py --model_id mpoyraz/wav2vec2-xls-r-300m-cv6-turkish --dataset speech-recognition-community-v2/dev_data --config tr --split validation --chunk_length_s 5.0 --stride_length_s 1.0

Evaluation results:

Dataset	WER	CER
Common Voice 6.1 TR test split	8.83	2.37
Speech Recognition Community dev data	32.81	11.22