wav2vec2_xlsr50k_korean_phoneme_aihub-40m

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on 40 minutes of the Korean Spontaneous Conversational speech dataset.

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

• Training data: AIHUB 자유대화 음성(일반남녀) dataset
• Num of speakers for training: 2
• Num of audio files: 425
• Total audio length: about 41 minutes
• No evaluation (will be updated soon)

Training procedure

• Token preprocessing: convert hangeul-jamo to MFA-v1-based phoneset (no longer available; v2 is based on ipa-format)
• Vocab size: 46

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0001
• train_batch_size: 8
• 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: 1000
• num_epochs: 200

Training results

Framework versions

• Transformers 4.21.3
• Pytorch 1.8.1+cu102
• Datasets 2.4.0
• Tokenizers 0.12.1