zhdoka

bori-asr

Deploy Dedicated

Dedicated Endpoints

Run this model inference on single tenant GPU with unmatched speed and reliability at scale.

Learn more

Container

Run this model inference with full control and performance in your environment.

Learn more

Get help setting up a custom Dedicated Endpoints.

Talk with our engineer to get a quote for reserved GPU instances with discounts.

README

License: apache-2.0

Metrics (500-sample eval, greedy)

CER: 3.84%
WER: 17.5%

Usage

python
import torch, librosa, numpy as np
from transformers import WhisperForConditionalGeneration, WhisperProcessor
m = WhisperForConditionalGeneration.from_pretrained('zhdoka/bori-asr').eval()
p = WhisperProcessor.from_pretrained('zhdoka/bori-asr')
a,_ = librosa.load('audio.wav', sr=16000, mono=True); a = a/ (np.abs(a).max() or 1); a = a[:30*16000]
feat = p(a, sampling_rate=16000, return_tensors='pt').input_features
ids = m.generate(feat, language='kazakh', task='transcribe', num_beams=1, max_new_tokens=225)
print(p.batch_decode(ids, skip_special_tokens=True)[0].lower().strip())