Whisper Small Urdu v1.0
This is a fine-tuned Whisper model for Urdu speech recognition.
Usage
1from transformers import AutoModelForSeq2SeqLM, AutoTokenizer
2
3model = AutoModelForSeq2SeqLM.from_pretrained("username/whisper-small-urdu_v1.0")
4tokenizer = AutoTokenizer.from_pretrained("username/whisper-small-urdu_v1.0")
5
6<!-- Here is the Code for Using-->
7
8
python exmaple
from transformers import WhisperForConditionalGeneration, WhisperProcessor
import torchaudio, torch
Reload merged model + processor
model = WhisperForConditionalGeneration.from_pretrained("/content/drive/MyDrive/FYP Models/whisper-small-urdu_v1.0")
processor = WhisperProcessor.from_pretrained("/content/drive/MyDrive/FYP Models/whisper-small-urdu_v1.0")
Load audio
waveform, sr = torchaudio.load("/content/drive/MyDrive/_سیب اور آم بازار می.m4a")
if waveform.shape[0] > 1:
waveform = waveform.mean(dim=0, keepdim=True)
if sr != 16000:
resampler = torchaudio.transforms.Resample(sr, 16000)
waveform = resampler(waveform)
audio_array = waveform.squeeze().numpy()
Prepare input
inputs = processor(audio_array, sampling_rate=16000, return_tensors="pt")
input_features = inputs.input_features
Run inference
with torch.no_grad():
pred_ids = model.generate(input_features)
pred_text = processor.batch_decode(pred_ids, skip_special_tokens=True)[0]
print("Pred:", pred_text)
