Downloads · 30 days
13
8% of all-time downloads
w2v-bert-2.0-bashkort-russian-omnivoice
AigizK/w2v-bert-2.0-bashkort-russian-omnivoice
w2v-bert-2.0-bashkort-russian-omnivoice is a automatic speech recognition model from AigizK. Use it when you need speech turned into text. The card lists the license as cc-by-4.0.
Модель для распознавания башкирской и русской речи на основе facebook/w2v-bert-2.0.
All-time downloads
161
Public
Parameters
606M
2.4 GB on disk
Likes
1
Public
Repo makeup
Click a slice to open those files.
.safetensors2.4 GB · 100%
At a glance
- Task
- Automatic Speech Recognition
- License
- cc-by-4.0
- Model type
- wav2vec2-bert
- Access
- Public
- Created
- Apr 12, 2026
- Updated
- Apr 18, 2026
- SHA
- e4c42789
Base models
Model card
From the Hugging Face model README
W2V-BERT 2.0 — Башкирский + Русский ASR
Модель для распознавания башкирской и русской речи на основе facebook/w2v-bert-2.0.
Данные для обучения
| Датасет | Язык | Описание |
|---|---|---|
| AigizK/bashkort_voice | Башкирский | Синтетический аудиодатасет башкирского языка, сгенерированный с помощью OmniVoice |
| bond005/sova_rudevices | Русский | — |
Результаты
Оценка проводилась на CommonVoice 25.0 (башкирский):
| Метрика | Значение |
|---|---|
| WER | 0.1523 |
Использование
from transformers import Wav2Vec2BertForCTC, Wav2Vec2BertProcessor
import torch
processor = Wav2Vec2BertProcessor.from_pretrained("AigizK/w2v-bert-2.0-bashkort-russian-omnivoice")
model = Wav2Vec2BertForCTC.from_pretrained("AigizK/w2v-bert-2.0-bashkort-russian-omnivoice")
inputs = processor(audio_array, sampling_rate=16000, return_tensors="pt")
with torch.no_grad():
logits = model(**inputs).logits
predicted_ids = torch.argmax(logits, dim=-1)
transcription = processor.batch_decode(predicted_ids)
- Task
- Automatic Speech Recognition
- Type
- wav2vec2-bert
- License
- cc-by-4.0
- Languages
- ba, ru
- Created
- Apr 12, 2026
- Updated
- Apr 18, 2026