wav2vec2-large-xls-r-300m-ur-cv8
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the common_voice dataset. It achieves the following results on the evaluation set:
- Loss: 1.1443
- Wer: 0.5677
Model description
More information needed
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: 0.0001
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 32
- 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
Training Loss | Epoch | Step | Validation Loss | Wer |
---|---|---|---|---|
3.6269 | 15.98 | 400 | 3.3246 | 1.0 |
3.0546 | 31.98 | 800 | 2.8148 | 0.9963 |
1.4589 | 47.98 | 1200 | 1.0237 | 0.6584 |
1.0911 | 63.98 | 1600 | 0.9524 | 0.5966 |
0.8879 | 79.98 | 2000 | 0.9827 | 0.5822 |
0.7467 | 95.98 | 2400 | 0.9923 | 0.5840 |
0.6427 | 111.98 | 2800 | 0.9988 | 0.5714 |
0.5685 | 127.98 | 3200 | 1.0872 | 0.5807 |
0.5068 | 143.98 | 3600 | 1.1194 | 0.5822 |
0.463 | 159.98 | 4000 | 1.1138 | 0.5692 |
0.4212 | 175.98 | 4400 | 1.1232 | 0.5714 |
0.4056 | 191.98 | 4800 | 1.1443 | 0.5677 |
Framework versions
- Transformers 4.16.0
- Pytorch 1.10.0+cu111
- Datasets 1.18.1
- Tokenizers 0.11.0
Evaluation Commands
- To evaluate on
mozilla-foundation/common_voice_8_0
with splittest
python eval.py --model_id anuragshas/wav2vec2-large-xls-r-300m-ur-cv8 --dataset mozilla-foundation/common_voice_8_0 --config ur --split test
Inference With LM
import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F
model_id = "anuragshas/wav2vec2-large-xls-r-300m-ur-cv8"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "ur", split="test", streaming=True, use_auth_token=True))
sample = next(sample_iter)
resampled_audio = F.resample(torch.tensor(sample["audio"]["array"]), 48_000, 16_000).numpy()
model = AutoModelForCTC.from_pretrained(model_id)
processor = AutoProcessor.from_pretrained(model_id)
input_values = processor(resampled_audio, return_tensors="pt").input_values
with torch.no_grad():
logits = model(input_values).logits
transcription = processor.batch_decode(logits.numpy()).text
# => "اب نے ٹ پیس ان لیتے ہیں"
Eval results on Common Voice 8 "test" (WER):
Without LM | With LM (run ./eval.py ) |
---|---|
52.146 | 42.376 |
- Downloads last month
- 13
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social
visibility and check back later, or deploy to Inference Endpoints (dedicated)
instead.
Dataset used to train anuragshas/wav2vec2-large-xls-r-300m-ur-cv8
Evaluation results
- Test WER on Common Voice 8self-reported42.376
- Test CER on Common Voice 8self-reported18.180