app.py

import os
import time
import asyncio
from fastapi import FastAPI
import edge_tts
from fastapi.responses import FileResponse
from pydub import AudioSegment

app = FastAPI()

# 🔹 Function to split text into manageable chunks
def split_text(text, max_chunk_size=500):
    """Split text into smaller chunks at sentence boundaries."""
    sentences = text.replace('।', '.').replace('؟', '?').split('.')
    chunks = []
    current_chunk = []
    current_length = 0

    for sentence in sentences:
        sentence = sentence.strip() + '.'
        sentence_length = len(sentence)

        if current_length + sentence_length > max_chunk_size and current_chunk:
            chunks.append(' '.join(current_chunk))
            current_chunk = []
            current_length = 0

        current_chunk.append(sentence)
        current_length += sentence_length

    if current_chunk:
        chunks.append(' '.join(current_chunk))

    return chunks

# 🔹 Function to process a single chunk asynchronously
async def process_chunk(text, voice, temp_dir, chunk_index):
    """Generate speech for a single chunk and save as MP3."""
    tmp_path = os.path.join(temp_dir, f"chunk_{chunk_index}.mp3")
    print(f"🎤 Processing chunk {chunk_index}: {text[:50]}...")  # Logging for debugging
    communicate = edge_tts.Communicate(text, voice)
    await communicate.save(tmp_path)
    return tmp_path

# 🔹 Function to merge all chunked MP3 files into a single audio file
async def combine_audio_files(chunk_files, output_path):
    """Combine multiple MP3 files into one final MP3."""
    combined = AudioSegment.empty()

    for file in chunk_files:
        print(f"🔹 Adding {file} to final output")  # Logging for debugging
        combined += AudioSegment.from_mp3(file)

    combined.export(output_path, format="mp3")

    # Remove temporary files
    for file in chunk_files:
        os.remove(file)

@app.get("/")
def home():
    return {"message": "✅ EdgeTTS FastAPI is running!"}

# 🔹 Main TTS API
@app.get("/tts")
async def tts(text: str, voice: str = "en-US-AriaNeural"):
    """Main API function to handle text-to-speech conversion."""
    temp_dir = "temp_audio"
    os.makedirs(temp_dir, exist_ok=True)

    chunks = split_text(text)

    # If text is short, process directly
    if len(chunks) == 1:
        print("📢 Processing without chunking...")
        output_file = await process_chunk(text, voice, temp_dir, 0)
        return FileResponse(output_file, media_type="audio/mpeg", filename="speech.mp3")

    print(f"🚀 Splitting into {len(chunks)} chunks and processing concurrently...")

    # 🔹 Concurrently process all chunks
    chunk_files = await asyncio.gather(*[
        process_chunk(ch, voice, temp_dir, i) for i, ch in enumerate(chunks)
    ])

    # 🔹 Merge all MP3 files
    output_file = "final_output.mp3"
    await combine_audio_files(chunk_files, output_file)

    print("✅ TTS Generation Complete. Sending response...")
    return FileResponse(output_file, media_type="audio/mpeg", filename="speech.mp3")

# 🔹 Ensure app starts in Hugging Face Spaces
if __name__ == "__main__":
    import uvicorn
    uvicorn.run(app, host="0.0.0.0", port=7860)