Vaani/_1.1_Audio.py

# !pip install huggingface_hub
# !pip install datasets


import os
import subprocess
import numpy as np
from scipy.io.wavfile import write
from tqdm import tqdm, trange
from collections import Counter
from datasets import load_dataset
from datasets import get_dataset_config_names


def array_to_wav(audio_array: np.ndarray, sample_rate: int, file_name: str):
    # Normalize if outside [-1, 1] to prevent clipping
    # if np.max(np.abs(audio_array)) > 1.0:
    #     audio_array = audio_array / np.max(np.abs(audio_array))
    # Convert to 16-bit PCM
    # audio_int16 = np.int16(audio_array * 32767)
    # Write to WAV file
    # write(file_name, sample_rate, audio_int16)
    write(file_name, sample_rate, audio_array)


configs = get_dataset_config_names("ARTPARK-IISc/Vaani")
english_district = [
    'Bellary', 'Bijapur', 'Churu', 'Etah', 'Guntur', 'JyotibaPhuleNagar',
    'Karimnagar', 'Krishna', 'NorthSouthGoa', 'Rajnandgaon', 'Sahebganj'
]
eng_dis_ds = [i for i in configs if i.split("_")[-1] in english_district]
print(len(eng_dis_ds), eng_dis_ds, "\n\n")


# HF_TOKEN = 
df_dict = {"audio_path":[], 'referenceImage':[], 'gender':[], 'state':[], 'district':[]}


for i, district in enumerate(eng_dis_ds):
    # district = "AndhraPradesh_Krishna"
    print("-"*2, i+1, "-"*10, district, "-"*10)

    cache_dir = f'/scratch/23m1521/Vaani/Audio-Cache/English/{district}'
    os.makedirs(cache_dir, exist_ok=True)
    
    ds = load_dataset("ARTPARK-IISc/Vaani", district, split="train", num_proc=10,
                      token=HF_TOKEN, cache_dir=cache_dir, streaming=False)
    english_rows = ds.filter(lambda x: x["language"] == "English")
    
    save_dir = f'/scratch/23m1521/Vaani/Audio/English/{district}'
    os.makedirs(save_dir, exist_ok=True)
    for i in tqdm(english_rows, dynamic_ncols=True, colour='blue'):
    # for i in ds:
        # if i["language"] == 'English':
        array_to_wav(
            i['audio']['array'],
            i['audio']['sampling_rate'], 
            os.path.join(save_dir, i['audio']['path'])
        )
        df_dict['audio_path'].append(i['audio']['path'])
        df_dict['referenceImage'].append(i['referenceImage'])
        df_dict['gender'].append(i['gender'])
        df_dict['state'].append(i['state'])
        df_dict['district'].append(i['district'])

    subprocess.run(["rm", "-rf", cache_dir], check=True)
    subprocess.run(["rm", "-rf", "/home/23m1521/.cache/huggingface/hub"], check=True)
    # break

import pandas as pd
pd.DataFrame(df_dict).to_csv("Vaani-Audio-Image-English.csv", index=False)