wav2vec2-large-xlsr-53-punjabi

This model is a fine-tuned version of Harveenchadha/vakyansh-wav2vec2-punjabi-pam-10 on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 1.2101
• Wer: 0.4939
• Cer: 0.2238

Evaluation Commands

1. To evaluate on mozilla-foundation/common_voice_8_0 with split test

python eval.py --model_id kingabzpro/wav2vec2-large-xlsr-53-punjabi --dataset mozilla-foundation/common_voice_8_0 --config pa-IN --split test

Inference With LM

import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F
model_id = "kingabzpro/wav2vec2-large-xlsr-53-punjabi"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "pa-IN", 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

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0003
• 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: 200
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
11.0563	3.7	100	1.9492	0.7123	0.3872
1.6715	7.41	200	1.3142	0.6433	0.3086
0.9117	11.11	300	1.2733	0.5657	0.2627
0.666	14.81	400	1.2730	0.5598	0.2534
0.4225	18.52	500	1.2548	0.5300	0.2399
0.3209	22.22	600	1.2166	0.5229	0.2372
0.2678	25.93	700	1.1795	0.5041	0.2276
0.2088	29.63	800	1.2101	0.4939	0.2238

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0