metadata

license: cc-by-4.0
language: tr
tags:
  - automatic-speech-recognition
  - mozilla-foundation/common_voice_7_0
  - tr
  - robust-speech-event
datasets:
  - mozilla-foundation/common_voice_7_0

wav2vec2-xls-r-300m-cv7-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 7.0 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 0.05
feat_proj_dropout 0.05
attention_dropout 0.05
final_dropout 0.05
activation_dropout 0.05
per_device_train_batch_size 8
per_device_eval_batch_size 8
gradient_accumulation_steps 8

Framework versions

Transformers 4.16.0.dev0
Pytorch 1.10.1
Datasets 1.17.0
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.