app.py

import torch
import torchaudio
import numpy as np
import gradio as gr
from f5_tts.api import F5TTS
from f5_tts.model.utils import convert_char_to_pinyin
from f5_tts.infer.utils_infer import transcribe

# Initialize F5TTS
tts = F5TTS()

# Load the model (Ensure correct file path)
try:
    tts.load_ema_model("hindi_tts_checkpoint.bin")  # Ensure correct path to model
except Exception as e:
    print(f"Error loading model: {e}")
    
# Function to generate speech
def generate_speech(text):
    """
    Convert Hindi text to speech and return the generated audio.
    """
    ref_audio_path = "E:/tts new/ref_audio.wav"  # Full path to ref_audio.wav (correct the path)
    ref_text = "यह संदर्भ ऑडियो का सामग्री, उपशीर्षक या लिप्यंतरण है।"  # Reference text

    # Ensure text is a single string (not a list)
    if isinstance(text, list):
        text = ' '.join(text)  # Join list elements to form a single string

    # Convert Hindi text to pinyin (if needed)
    pinyin_text = convert_char_to_pinyin(text)

    # Flatten any nested lists in pinyin_text
    if isinstance(pinyin_text, list):
        pinyin_text = [item for sublist in pinyin_text for item in (sublist if isinstance(sublist, list) else [sublist])]

    # Now, join the flattened list into a string
    pinyin_text = ' '.join(pinyin_text)

    # Generate speech
    try:
        output_waveform, sampling_rate = tts.infer(ref_audio_path, ref_text, pinyin_text)
        output_waveform = output_waveform.numpy()  # Convert to NumPy format
        return (sampling_rate, output_waveform)
    except Exception as e:
        return f"Error in speech generation: {e}"

# Gradio Interface
iface = gr.Interface(
    fn=generate_speech,
    inputs=gr.Textbox(label="Enter Hindi Text"),
    outputs=gr.Audio(label="Generated Speech"),
    title="Hindi Text-to-Speech (TTS)",
    description="Enter Hindi text, and the model will generate a speech output.",
)

# Launch the Gradio app
if __name__ == "__main__":
    iface.launch()