Downloads · 30 days
60
6% of all-time downloads
anuragshas/wav2vec2-xls-r-300m-sk-cv8-with-lm
wav2vec2-xls-r-300m-sk-cv8-with-lm 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
60
6% of all-time downloads
All-time downloads
1K
Public
Repo size
32.1 GB
Likes
0
Public
Click a slice to open those files.
.bin1.8 GB · 100%
From the Hugging Face model README
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - SK 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 |
|---|---|---|---|---|
| 5.175 | 2.41 | 400 | 4.6909 | 1.0 |
| 3.3785 | 4.82 | 800 | 3.3080 | 1.0 |
| 2.6964 | 7.23 | 1200 | 2.0651 | 1.1055 |
| 1.3008 | 9.64 | 1600 | 0.5845 | 0.6207 |
| 1.1185 | 12.05 | 2000 | 0.4195 | 0.4193 |
| 1.0252 | 14.46 | 2400 | 0.3824 | 0.3570 |
| 0.935 | 16.87 | 2800 | 0.3693 | 0.3462 |
| 0.8818 | 19.28 | 3200 | 0.3587 | 0.3318 |
| 0.8534 | 21.69 | 3600 | 0.3420 | 0.3180 |
| 0.8137 | 24.1 | 4000 | 0.3426 | 0.3130 |
| 0.7968 | 26.51 | 4400 | 0.3349 | 0.3102 |
| 0.7558 | 28.92 | 4800 | 0.3216 | 0.3019 |
| 0.7313 | 31.33 | 5200 | 0.3451 | 0.3060 |
| 0.7358 | 33.73 | 5600 | 0.3272 | 0.2967 |
| 0.718 | 36.14 | 6000 | 0.3315 | 0.2882 |
| 0.6991 | 38.55 | 6400 | 0.3299 | 0.2830 |
| 0.6529 | 40.96 | 6800 | 0.3140 | 0.2836 |
| 0.6225 | 43.37 | 7200 | 0.3128 | 0.2751 |
| 0.633 | 45.78 | 7600 | 0.3211 | 0.2774 |
| 0.5876 | 48.19 | 8000 | 0.3162 | 0.2764 |
| 0.588 | 50.6 | 8400 | 0.3082 | 0.2722 |
| 0.5915 | 53.01 | 8800 | 0.3120 | 0.2681 |
| 0.5798 | 55.42 | 9200 | 0.3133 | 0.2709 |
| 0.5736 | 57.83 | 9600 | 0.3086 | 0.2676 |
mozilla-foundation/common_voice_8_0 with split testpython eval.py --model_id anuragshas/wav2vec2-xls-r-300m-sk-cv8-with-lm --dataset mozilla-foundation/common_voice_8_0 --config sk --split test
speech-recognition-community-v2/dev_datapython eval.py --model_id anuragshas/wav2vec2-xls-r-300m-sk-cv8-with-lm --dataset speech-recognition-community-v2/dev_data --config sk --split validation --chunk_length_s 5.0 --stride_length_s 1.0
import torch
from datasets import load_dataset
from transformers import AutoModelForCTC, AutoProcessor
import torchaudio.functional as F
model_id = "anuragshas/wav2vec2-xls-r-300m-sk-cv8-with-lm"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "sk", 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
# => ""
| Without LM | With LM (run ./eval.py) |
|---|---|
| 26.707 | 18.609 |