Wav2vec2 Xls R 300m Cv6 Turkish

🚀 wav2vec2-xls-r-300m-cv6-turkish

This is an Automatic Speech Recognition (ASR) model fine - tuned on the Turkish language, offering high - quality speech recognition capabilities.

🚀 Quick Start

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

✨ Features

• Fine - tuned on Turkish language for accurate Automatic Speech Recognition.
• Supports multiple datasets for training and evaluation.
• Utilizes an N - gram language model trained on Turkish Wikipedia articles.

📦 Installation

Before running evaluation, please install the unicode_tr package. It is used for Turkish text processing.

💻 Usage Examples

Basic Usage

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

Advanced Usage

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

📚 Documentation

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 were performed and wav2vec2-turkish repo was used for that purpose.

Training hyperparameters

The following hyperparameters 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 results

🔧 Technical Details

This model is based on the fine - tuning of facebook/wav2vec2-xls-r-300m on Turkish language. Custom pre - processing and loading steps were implemented to support multiple datasets. Hyperparameters were carefully tuned to achieve good performance on Turkish speech recognition tasks.

📄 License

This project is licensed under the Apache - 2.0 license.