Create main.py

main.py

@@ -0,0 +1,75 @@
+import os
+import re
+import io
+import numpy as np
+import scipy.io.wavfile as wavfile
+from fastapi import FastAPI, Depends, HTTPException, Security
+from fastapi.security.api_key import APIKeyHeader
+from fastapi.responses import Response
+from pydantic import BaseModel
+
+# Kokoro imports
+from kokoro import KPipeline
+
+app = FastAPI(title="Kokoro-82M TTS Backend")
+
+# Security setup
+API_KEY_NAME = "X-API-KEY"
+api_key_header = APIKeyHeader(name=API_KEY_NAME, auto_error=False)
+
+def get_api_key(api_key: str = Security(api_key_header)):
+    expected_key = os.environ.get("SECRET_API_KEY")
+    if not expected_key or api_key != expected_key:
+        raise HTTPException(status_code=403, detail="Invalid or missing API Key")
+    return api_key
+
+# Singleton Model Loading
+# Loads 'a' for American English. Change to 'b' for British.
+pipeline = None
+
+@app.on_event("startup")
+def load_model():
+    global pipeline
+    # Initialize the model once into memory
+    pipeline = KPipeline(lang_code='a')
+
+class TTSRequest(BaseModel):
+    text: str
+    voice_id: str = "af_bella"
+
+def chunk_text(text: str):
+    """Splits text by punctuation to avoid Kokoro's context limit."""
+    # Split by ., !, ? followed by a space
+    chunks = re.split(r'(?<=[.!?]) +', text.strip())
+    return [c for c in chunks if c.strip()]
+
+@app.post("/generate", dependencies=[Depends(get_api_key)])
+async def generate_audio(request: TTSRequest):
+    try:
+        chunks = chunk_text(request.text)
+        audio_pieces = []
+        sample_rate = 24000 # Kokoro default
+
+        # Process chunks sequentially
+        for chunk in chunks:
+            # pipeline returns a generator of (graphemes, phonemes, audio)
+            generator = pipeline(chunk, voice=request.voice_id, speed=1.0, split_pattern=None)
+            for _, _, audio in generator:
+                if audio is not None:
+                    audio_pieces.append(audio)
+
+        if not audio_pieces:
+            raise HTTPException(status_code=400, detail="Could not generate audio from input text.")
+
+        # Concatenate all numpy audio chunks
+        final_audio = np.concatenate(audio_pieces)
+
+        # Convert to standard WAV bytes in memory
+        wav_io = io.BytesIO()
+        wavfile.write(wav_io, sample_rate, final_audio)
+        wav_io.seek(0)
+
+        return Response(content=wav_io.read(), media_type="audio/wav")
+
+    except Exception as e:
+        raise HTTPException(status_code=500, detail=str(e))