Model reference · open weights
wav2vec2-large-xls-r-bg is an open-weight audio or speech model from anuragshas. AxForge deploys and operates it for you on dedicated EU-owned hardware — with the licence handled where one is required.
Available as managed deployment — configured and operated for you on dedicated EU hardware, quoted per deployment.
What it is
| Released by | anuragshas |
|---|---|
| Type | Audio & music |
| Task | Speech→text |
| Runs with | transformers |
| Released | 2022-03-02 |
| Popularity | 1k downloads / month |
| Licence | Open weights |
About
This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - BG 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 |
|---|---|---|---|---|
| 3.1589 | 3.48 | 400 | 3.0830 | 1.0 |
| 2.8921 | 6.96 | 800 | 2.6605 | 0.9982 |
| 1.3049 | 10.43 | 1200 | 0.5069 | 0.5707 |
| 1.1349 | 13.91 | 1600 | 0.4159 | 0.5041 |
| 1.0686 | 17.39 | 2000 | 0.3815 | 0.4746 |
| 0.999 | 20.87 | 2400 | 0.3541 | 0.4343 |
| 0.945 | 24.35 | 2800 | 0.3266 | 0.4132 |
| 0.9058 | 27.83 | 3200 | 0.2969 | 0.3771 |
| 0.8672 | 31.3 | 3600 | 0.2802 | 0.3553 |
| 0.8313 | 34.78 | 4000 | 0.2662 | 0.3380 |
| 0.8068 | 38.26 | 4400 | 0.2528 | 0.3181 |
| 0.7796 | 41.74 | 4800 | 0.2537 | 0.3073 |
| 0.7621 | 45.22 | 5200 | 0.2503 | 0.3036 |
| 0.7611 | 48.7 | 5600 | 0.2477 | 0.2991 |
mozilla-foundation/common_voice_8_0 with split testpython eval.py --model_id anuragshas/wav2vec2-large-xls-r-300m-bg --dataset mozilla-foundation/common_voice_8_0 --config bg --split test
speech-recognition-community-v2/dev_datapython eval.py --model_id anuragshas/wav2vec2-large-xls-r-300m-bg --dataset speech-recognition-community-v2/dev_data --config bg --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-large-xls-r-300m-bg"
sample_iter = iter(load_dataset("mozilla-foundation/common_voice_8_0", "bg", 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
# => "и надутият му ката блоонкурем взе да се събира"
| Without LM | With LM (run ./eval.py) |
|---|---|
| 30.07 | 21.195 |
From the published model card. Full card on the HuggingFace links in the sidebar.
Benchmarks
As published on the model card — the maker's own numbers, not measured by AxForge.
| Task | Dataset | Metric | Score |
|---|---|---|---|
| Automatic Speech Recognition | Common Voice 8 | Test WER | 21.195 |
| Automatic Speech Recognition | Common Voice 8 | Test CER | 4.786 |
| Automatic Speech Recognition | Robust Speech Event - Dev Data | Test WER | 32.667 |
| Automatic Speech Recognition | Robust Speech Event - Dev Data | Test CER | 12.452 |
| Automatic Speech Recognition | Robust Speech Event - Test Data | Test WER | 31.030 |
Using it via the API
Once AxForge deploys wav2vec2-large-xls-r-bg for you, it answers on the OpenAI-compatible API — the same base URL and keys as every other model. (wav2vec2-large-xls-r-bg below is illustrative; you get the exact model name on deployment.)
$ curl -sS https://api.axforge.ai/v1/audio/transcriptions \ -H "Authorization: Bearer $AXFORGE_API_KEY" \ -F model="wav2vec2-large-xls-r-bg" -F file=@audio.mp3
Create an account — your API key is available in the console. 3M free tokens every 30 days with every new account.