app.py

import librosa
import numpy as np

def preprocess_audio(file_path):
    y, sr = librosa.load(file_path, sr=16000)
    mel_spectrogram = librosa.feature.melspectrogram(y=y, sr=sr)
    return mel_spectrogram

from transformers import Wav2Vec2ForCTC, Wav2Vec2Tokenizer, MarianMTModel, MarianTokenizer

# Load pre-trained models
speech_to_text_model = Wav2Vec2ForCTC.from_pretrained("facebook/wav2vec2-large-xlsr-53")
speech_to_text_tokenizer = Wav2Vec2Tokenizer.from_pretrained("facebook/wav2vec2-large-xlsr-53")

translation_model = MarianMTModel.from_pretrained("Helsinki-NLP/opus-mt-en-hi")
translation_tokenizer = MarianTokenizer.from_pretrained("Helsinki-NLP/opus-mt-en-hi")

def translate_audio(file_path):
    # Preprocess the audio
    mel_spectrogram = preprocess_audio(file_path)
    
    # Convert to text
    audio_input = speech_to_text_tokenizer(file_path, return_tensors="pt").input_values
    logits = speech_to_text_model(audio_input).logits
    predicted_ids = logits.argmax(dim=-1)
    transcription = speech_to_text_tokenizer.batch_decode(predicted_ids)[0]
    
    # Translate text
    translation_input = translation_tokenizer(transcription, return_tensors="pt")
    translated_output = translation_model.generate(**translation_input)
    translation = translation_tokenizer.batch_decode(translated_output, skip_special_tokens=True)[0]
    
    return translation

import datetime

def should_translate():
    now = datetime.datetime.now()
    return now.hour >= 18

def handle_translation(file_path):
    if should_translate():
        return translate_audio(file_path)
    else:
        return "Translation is only available after 6 PM IST."