mms-common_voice_13_0-eo-1

mms-common_voice_13_0-eo-1, an Esperanto speech recognizer

This model is a fine-tuned version of patrickvonplaten/mms-300m on the the mozilla-foundation/common_voice_13_0 Esperanto dataset. It achieves the following results on the evaluation set:

• Loss: 0.2257
• Cer: 0.0209
• Wer: 0.0678

While the training loss is lower, this model does not perform significantly better than xekri/wav2vec2-common_voice_13_0-eo-3.

The first 10 samples in the test set:

Model description

See patrickvonplaten/mms-300m, or equivalently, facebook/wav2vec2-large-xlsr-53, as it seems to me that the only difference is that the speech front-end was trained with more languages and data in the mms-300m checkpoint.

Intended uses & limitations

Speech recognition for Esperanto. The base model was pretrained and finetuned on 16kHz sampled speech audio. When using the model make sure that your speech input is also sampled at 16KHz.

Training and evaluation data

The training split was set to train[:15000] while the eval split was set to validation[:1500].

Training procedure

The same as xekri/wav2vec2-common_voice_13_0-eo-3.

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 3e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• layerdrop: 0.1
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Framework versions

• Transformers 4.29.1
• Pytorch 2.0.1+cu118
• Datasets 2.12.0
• Tokenizers 0.13.3