File size: 1,762 Bytes
c4ef99f |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 |
# Arabic syllables recognition with tashkeel.
This is fine tuned wav2vec2 model to recognize arabic syllables from speech.
The model was trained on Modern standard arabic dataset.\
5-gram language model is available with the model.
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/Syllables_final_Large')
model = Wav2Vec2ForCTC.from_pretrained("IbrahimSalah/Syllables_final_Large")
```
```
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])
```
|