wav2vec2-bloom-speech-hbb

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 - HBB (Nya Huba) 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.

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

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

The following hyperparameters were used during training:

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
No log	14.69	250	0.5759	0.6270	0.1577
1.9895	29.4	500	0.4064	0.3713	0.0784
1.9895	44.11	750	0.4197	0.3755	0.0800
0.1048	58.8	1000	0.4609	0.3638	0.0824
0.1048	73.51	1250	0.4703	0.3747	0.0840
0.0554	88.23	1500	0.5218	0.3361	0.0798
0.0554	102.91	1750	0.5258	0.3328	0.0723
0.0399	117.63	2000	0.5095	0.3227	0.0728
0.0399	132.34	2250	0.4456	0.3621	0.0859
0.0335	147.06	2500	0.5449	0.3663	0.0824
0.0335	161.74	2750	0.4750	0.3127	0.0662
0.0275	176.46	3000	0.5114	0.3202	0.0679
0.0275	191.17	3250	0.5342	0.2892	0.0618
0.0208	205.86	3500	0.5613	0.3286	0.0681
0.0208	220.57	3750	0.5267	0.3051	0.0677
0.0174	235.29	4000	0.5336	0.2749	0.0620
0.0174	249.97	4250	0.5204	0.2867	0.0606
0.0192	264.69	4500	0.5137	0.2758	0.0613
0.0192	279.4	4750	0.4783	0.3060	0.0691