You can deploy this model and start earning money today!

KasuleTrevor/
cdli-whisper-en-ug-sunbird-full-a40

Name: cdli-whisper-en-ug-sunbird-full-a40
Availability: InStock

transformers

safetensors

whisper

automatic-speech-recognition

generated_from_trainer

cdli/ugandan_english_nonstandard_speech_v1.0

Sunbird/asr-whisper-large-v3-salt

finetune

mit

endpoints_compatible

Views

No views yet

cdli-whisper-en-ug-sunbird-full-a40

This model is a fine-tuned version of Sunbird/asr-whisper-large-v3-salt on the cdli/ugandan_english_nonstandard_speech_v1.0 dataset. It achieves the following results on the evaluation set:

Loss: 0.9112
Wer: 0.3162
Cer: 0.2244
test_cer = 0.1313
test_loss = 0.6542
test_runtime = 0:09:47.06
test_samples_per_second = 1.726
test_steps_per_second = 0.433
test_wer = 0.2122

Model description

This is a model that is trained on atypical speech for Ugandan English. The full model was finetuned.

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 2e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 4
total_train_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 150
training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.6153	0.7734	250	0.8760	0.3670	0.2720
0.449	1.5445	500	0.8358	0.3330	0.2422
0.377	2.3155	750	0.8675	0.3321	0.2382
0.3099	3.0866	1000	0.8733	0.3141	0.2215
0.3111	3.8600	1250	0.8768	0.3126	0.2218
0.2705	4.6311	1500	0.9011	0.3128	0.2230
0.2695	5.4022	1750	0.9089	0.3148	0.2232
0.2369	6.1732	2000	0.9112	0.3162	0.2244

Framework versions

Transformers 4.52.0
Pytorch 2.7.1+cu118
Datasets 3.6.0
Tokenizers 0.21.4

KasuleTrevor/cdli-whisper-en-ug-sunbird-full-a40