app.py

# import streamlit as st
# from transformers import SeamlessM4Tv2Model, AutoProcessor
# import torch
# import numpy as np
# from scipy.io.wavfile import write
# import re
# from io import BytesIO

# # Load the processor and model
# processor = AutoProcessor.from_pretrained("facebook/seamless-m4t-v2-large")
# model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large")

# device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
# model.to(device)

# # Number to words function for Uzbek
# number_words = {
#     0: "nol", 1: "bir", 2: "ikki", 3: "uch", 4: "to'rt", 5: "besh", 6: "olti", 7: "yetti", 8: "sakkiz", 9: "to'qqiz",
#     10: "o'n", 11: "o'n bir", 12: "o'n ikki", 13: "o'n uch", 14: "o'n to'rt", 15: "o'n besh", 16: "o'n oltı", 17: "o'n yetti",
#     18: "o'n sakkiz", 19: "o'n toqqiz", 20: "yigirma", 30: "o'ttiz", 40: "qirq", 50: "ellik", 60: "oltmish", 70: "yetmish",
#     80: "sakson", 90: "to'qson", 100: "yuz", 1000: "ming", 1000000: "million"
# }

# def number_to_words(number):
#     if number < 20:
#         return number_words[number]
#     elif number < 100:
#         tens, unit = divmod(number, 10)
#         return number_words[tens * 10] + (" " + number_words[unit] if unit else "")
#     elif number < 1000:
#         hundreds, remainder = divmod(number, 100)
#         return (number_words[hundreds] + " yuz" if hundreds > 1 else "yuz") + (" " + number_to_words(remainder) if remainder else "")
#     elif number < 1000000:
#         thousands, remainder = divmod(number, 1000)
#         return (number_to_words(thousands) + " ming" if thousands > 1 else "ming") + (" " + number_to_words(remainder) if remainder else "")
#     elif number < 1000000000:
#         millions, remainder = divmod(number, 1000000)
#         return number_to_words(millions) + " million" + (" " + number_to_words(remainder) if remainder else "")
#     elif number < 1000000000000:
#         billions, remainder = divmod(number, 1000000000)
#         return number_to_words(billions) + " milliard" + (" " + number_to_words(remainder) if remainder else "")
#     else:
#         return str(number)

# def replace_numbers_with_words(text):
#     def replace(match):
#         number = int(match.group())
#         return number_to_words(number)
#     result = re.sub(r'\b\d+\b', replace, text)
#     return result

# # Replacements
# replacements = [
#     ("bo‘ladi", "bo'ladi"),
#     ("yog‘ingarchilik", "yog'ingarchilik"),
# ]

# def cleanup_text(text):
#     for src, dst in replacements:
#         text = text.replace(src, dst)
#     return text

# # Streamlit App
# st.title("Text-to-Speech using Seamless M4T Model")

# # User Input
# user_input = st.text_area("Enter the text for speech generation", height=200)

# # Process the text and generate speech
# if st.button("Generate Speech"):
#     if user_input.strip():
#         # Apply text transformations
#         converted_text = replace_numbers_with_words(user_input)
#         cleaned_text = cleanup_text(converted_text)
        
#         # Process input for model
#         inputs = processor(text=cleaned_text, src_lang="uzn", return_tensors="pt").to(device)
        
#         # Generate audio from text
#         audio_array_from_text = model.generate(**inputs, tgt_lang="uzn")[0].cpu().numpy().squeeze()

#         # Save to BytesIO
#         audio_io = BytesIO()
#         write(audio_io, 16000, audio_array_from_text.astype(np.float32))
#         audio_io.seek(0)

#         # Provide audio for playback
#         st.audio(audio_io, format='audio/wav')
#     else:
#         st.warning("Please enter some text to generate speech.")
import streamlit as st
from transformers import SeamlessM4TTokenizer, SeamlessM4Tv2Model
import torch
import numpy as np
from scipy.io.wavfile import write
from io import BytesIO

# Load the tokenizer and model
# tokenizer = SeamlessM4TTokenizer.from_pretrained("facebook/seamless-m4t-v2-large")
# model = SeamlessM4Tv2Model.from_pretrained("facebook/seamless-m4t-v2-large")

# Load model directly
from transformers import AutoProcessor, AutoModelForTextToSpectrogram

processor = AutoProcessor.from_pretrained("Beehzod/speecht5_finetuned_uz_customData")
model = AutoModelForTextToSpectrogram.from_pretrained("Beehzod/speecht5_finetuned_uz_customData")

# Set the device (CUDA if available, else CPU)
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)

# Streamlit title
st.title("Text-to-Speech with Seamless M4T Model")

# Input text field
text = st.text_area("Enter text for audio generation")

# Button to generate audio
if st.button("Generate Audio"):
    if text:
        # Preprocess the text and convert to tensor
        inputs = tokenizer(text=text, src_lang="uzn", return_tensors="pt").to(device)

        # Generate audio from the model
        audio_array_from_text = model.generate(**inputs, tgt_lang="uzn")[0].cpu().numpy().squeeze()

        # Save the audio as a .wav file in memory
        audio_file = BytesIO()
        write(audio_file, 16000, audio_array_from_text.astype(np.float32))
        audio_file.seek(0)  # Reset the pointer to the start of the file

        # Display the audio player in the Streamlit app
        st.audio(audio_file, format="audio/wav")
    else:
        st.warning("Please enter text to generate audio.")