Downloads · 30 days
30
2% of all-time downloads
anuragshas/wav2vec2-large-xls-r-300m-ha-cv8
wav2vec2-large-xls-r-300m-ha-cv8 is a automatic speech recognition model from anuragshas. Use it when you need speech turned into text. It is set up for transformers. The card lists the license as apache-2.0.
should probably proofread and complete it, then remove this comment. --
Downloads · 30 days
30
2% of all-time downloads
All-time downloads
1.7K
Public
Repo size
5.6 GB
Likes
1
Public
Click a slice to open those files.
.bin3 GB · 100%
From the Hugging Face model README
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:
More information needed
More information needed
More information needed
The following hyperparameters were used during training:
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 2.9599 | 6.56 | 400 | 2.8650 | 1.0 |
| 2.7357 | 13.11 | 800 | 2.7377 | 0.9951 |
| 1.3012 | 19.67 | 1200 | 0.6686 | 0.7111 |
| 1.0454 | 26.23 | 1600 | 0.5686 | 0.6137 |
| 0.9069 | 32.79 | 2000 | 0.5576 | 0.5815 |
| 0.82 | 39.34 | 2400 | 0.5502 | 0.5591 |
| 0.7413 | 45.9 | 2800 | 0.5970 | 0.5586 |
| 0.6872 | 52.46 | 3200 | 0.5817 | 0.5428 |
| 0.634 | 59.02 | 3600 | 0.5636 | 0.5314 |
| 0.6022 | 65.57 | 4000 | 0.5780 | 0.5229 |
| 0.5705 | 72.13 | 4400 | 0.6036 | 0.5323 |
| 0.5408 | 78.69 | 4800 | 0.6119 | 0.5336 |
| 0.5225 | 85.25 | 5200 | 0.6105 | 0.5270 |
| 0.5265 | 91.8 | 5600 | 0.6034 | 0.5231 |
| 0.5154 | 98.36 | 6000 | 0.6094 | 0.5234 |
mozilla-foundation/common_voice_8_0 with split testpython eval.py --model_id anuragshas/wav2vec2-large-xls-r-300m-ha-cv8 --dataset mozilla-foundation/common_voice_8_0 --config ha --split test
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-ha-cv8"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "ha", 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
# => "kakin hade ya ke da kyautar"
| Without LM | With LM (run ./eval.py) |
|---|---|
| 47.821 | 36.295 |