Create app.py

app.py

@@ -0,0 +1,64 @@
+import io
+import base64
+import torch
+import torchaudio
+from pydub import AudioSegment
+from chatterbox import mtl_tts
+from huggingface_hub import snapshot_download
+from safetensors.torch import load_file as load_safetensors
+from fastapi import FastAPI
+from fastapi.responses import JSONResponse
+from pydantic import BaseModel
+
+app = FastAPI()
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+print("⏳ Loading model...")
+
+ckpt_dir = snapshot_download(
+    repo_id="NAMAA-Space/NAMAA-Saudi-TTS",
+    repo_type="model"
+)
+
+model = mtl_tts.ChatterboxMultilingualTTS.from_pretrained(device=device)
+
+t3_state = load_safetensors(f"{ckpt_dir}/t3_mtl23ls_v2.safetensors", device=device)
+model.t3.load_state_dict(t3_state)
+
+print(f"✅ Model loaded on {device}")
+
+
+class TTSRequest(BaseModel):
+    text: str
+
+
+@app.post("/tts")
+def tts(req: TTSRequest):
+    try:
+        with torch.no_grad():
+            audio_tensor = model.generate(req.text)
+
+        audio_np = audio_tensor.cpu().numpy().squeeze()
+        sample_rate = 24000
+
+        audio_io = io.BytesIO()
+        torchaudio.save(
+            audio_io,
+            torch.from_numpy(audio_np).unsqueeze(0),
+            sample_rate,
+            format="wav"
+        )
+        audio_io.seek(0)
+
+        audio = AudioSegment.from_wav(audio_io)
+        mp3_io = io.BytesIO()
+        audio.export(mp3_io, format="mp3", bitrate="192k")
+        mp3_io.seek(0)
+
+        audio_base64 = base64.b64encode(mp3_io.read()).decode("utf-8")
+
+        return {"audio": audio_base64}
+
+    except Exception as e:
+        return JSONResponse({"error": str(e)}, status_code=500)