YAML Metadata
Warning:
empty or missing yaml metadata in repo card
(https://huggingface.co/docs/hub/model-cards#model-card-metadata)
Quran syllables recognition with tashkeel.
This is fine tuned wav2vec2 model to recognize quran syllables from speech.
The model was trained on Tarteel dataset after cleanning and converting into syllables .
5-gram language model is available with the model.
The model transcripe audio speech into syllables .
For instance, when presented with the audio and transcription "ู
ููู ุงููุฌููููุฉู ููุงููููุงุณู" the expected model output would be
"ู
ู ูููู ุฌููู ูู ุชู ูููู ููุงูุณู" .
To try it out :
!pip install datasets transformers
!pip install https://github.com/kpu/kenlm/archive/master.zip pyctcdecode
from transformers import Wav2Vec2Processor, Wav2Vec2ForCTC
from transformers import Wav2Vec2ProcessorWithLM
processor = Wav2Vec2ProcessorWithLM.from_pretrained('IbrahimSalah/Wav2vecXXl_quran_syllables')
model = Wav2Vec2ForCTC.from_pretrained("IbrahimSalah/Wav2vecXXl_quran_syllables")
import pandas as pd
dftest = pd.DataFrame(columns=['audio'])
import datasets
from datasets import Dataset
path ='/content/908-33.wav'
dftest['audio']=[path] ## audio path
dataset = Dataset.from_pandas(dftest)
import torch
import torchaudio
def speech_file_to_array_fn(batch):
speech_array, sampling_rate = torchaudio.load(batch["audio"])
print(sampling_rate)
resampler = torchaudio.transforms.Resample(sampling_rate, 16_000) # The original data was with 48,000 sampling rate. You can change it according to your input.
batch["audio"] = resampler(speech_array).squeeze().numpy()
return batch
import numpy as np
from datasets import load_dataset
from transformers import Wav2Vec2ForCTC, Wav2Vec2Processor
test_dataset = dataset.map(speech_file_to_array_fn)
inputs = processor(test_dataset["audio"], sampling_rate=16_000, return_tensors="pt", padding=True)
with torch.no_grad():
logits = model(inputs.input_values).logits
print(logits.numpy().shape)
transcription = processor.batch_decode(logits.numpy()).text
print("Prediction:",transcription[0])
sample audio 1-
2-
Output
1- ุกููู ููุงู ุกู ูู ููู
ู ูู ูููู ูููู ู
ูุนู ุฌู ุฒููู ูู ูููู ุกูุฑู ุถู ูู ู
ูุงู ููุงู ูู ูู ฺพูู
ู ู
ููู ุกููู ูู ููุงู
2- ุกูุฐู ููุงู ูู ูููู ุณู ูู ูู ุกููู ุจููู ฺพู ููุงู ุกู ุจู ุชู ุกููู ูููู ุฑู ุกููู ุชู ุกู ุญู ุฏู ุนู ุดู ุฑู ูููู ูู ุจููู ููุดู ุดูู
ู ุณู ูููู ูู ู
ู ุถู ุฑู ุกููู ุชู ฺพูู
ู ูููู ุณูุงู ุฌู ุฏููููู
- Downloads last month
- 534