automatic-speech-recognition sil-ai/bloom-speech generated_from_trainer

wav2vec2-bloom-speech-kek

logo for Bloom Library sil-ai logo

Model description

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the SIL-AI/bloom-speech - KEK (Chinese, Hakka) dataset. It achieves the following results on the evaluation set:

Users should refer to the original model for tutorials on using a trained model for inference.

Intended uses & limitations

Users of this model must abide by the SIL RAIL-M License.

This model is created as a proof of concept and no guarantees are made regarding the performance of the model is specific situations.

Training and evaluation data

Training, Validation, and Test datasets were generated from the same corpus, ensuring that no duplicate files were used.

Training procedure

Standard finetuning of XLS-R was used based on the examples in the Hugging Face Transformers Github

Training hyperparameters

The following hyperparameters were used during training:

Training results

Training Loss Epoch Step Validation Loss Wer Cer
No log 4.8 250 3.2004 1.0 1.0
5.5399 9.61 500 2.9360 1.0 1.0
5.5399 14.42 750 2.0897 1.0 0.8686
2.1248 19.23 1000 0.9416 0.8812 0.2877
2.1248 24.04 1250 0.6596 0.7981 0.2522
0.6897 28.84 1500 0.4984 0.6247 0.1798
0.6897 33.65 1750 0.4491 0.6247 0.1674
0.408 38.46 2000 0.4697 0.6722 0.1715
0.408 43.27 2250 0.4966 0.5962 0.1591
0.3001 48.08 2500 0.4297 0.5321 0.1452
0.3001 52.88 2750 0.4405 0.5534 0.1420
0.2305 57.69 3000 0.4567 0.5582 0.1512
0.2305 62.5 3250 0.4950 0.5487 0.1498

Framework versions