Create app.py

app.py

@@ -0,0 +1,73 @@
+import os
+from zipfile import ZipFile
+
+import torch
+from fastapi import FastAPI
+from fastapi.responses import StreamingResponse
+
+from OpenVoice import se_extractor  # لاحقاً لو حبيت تعمل voice cloning
+from OpenVoice.api import BaseSpeakerTTS, ToneColorConverter
+
+app = FastAPI()
+
+# -------- إعداد الموديل عند تشغيل الـ Space --------
+
+CHECKPOINT_ZIP_URL = "https://myshell-public-repo-hosting.s3.amazonaws.com/checkpoints_1226.zip"
+CKPT_DIR = "checkpoints"
+EN_CKPT_BASE = os.path.join(CKPT_DIR, "base_speakers", "EN")
+CONVERTER_CKPT = os.path.join(CKPT_DIR, "converter")
+
+os.makedirs("outputs", exist_ok=True)
+
+if not os.path.exists(CKPT_DIR):
+    print("Downloading OpenVoice checkpoints ...")
+    os.system(f"wget {CHECKPOINT_ZIP_URL} -O ckpt.zip")
+    print("Extracting checkpoints ...")
+    ZipFile("ckpt.zip").extractall()
+    print("Checkpoints ready.")
+
+device = "cuda" if torch.cuda.is_available() else "cpu"
+
+# Base TTS (إنجليزي أساساً – العربي عن طريق cross-lingual cloning)
+base_speaker_tts = BaseSpeakerTTS(f"{EN_CKPT_BASE}/config.json", device=device)
+base_speaker_tts.load_ckpt(f"{EN_CKPT_BASE}/checkpoint.pth")
+
+# لو لاحقاً حابب تعمل cloning لصوت معيّن:
+# tone_color_converter = ToneColorConverter(f"{CONVERTER_CKPT}/config.json", device=device)
+# tone_color_converter.load_ckpt(f"{CONVERTER_CKPT}/checkpoint.pth")
+
+
+# -------- API Endpoint --------
+
+@app.post("/tts")
+async def tts_endpoint(
+    text: str,
+    speaker: str = "default",  # ممكن تجرب: default, cheerful, sad, angry ...
+    speed: float = 1.0,
+):
+    """
+    يحوّل النص إلى صوت WAV باستخدام OpenVoice.
+    """
+    out_path = "outputs/out.wav"
+
+    # اللغة الافتراضية EN – لو هتكتب عربي ممكن يطلع لكن بجودة أقل
+    language = "English"
+
+    # دالة tts من BaseSpeakerTTS
+    base_speaker_tts.tts(
+        text=text,
+        output_path=out_path,
+        speaker=speaker,
+        language=language,
+        speed=speed,
+    )
+
+    audio_file = open(out_path, "rb")
+    return StreamingResponse(audio_file, media_type="audio/wav")
+
+
+# مهم جدًا لـ Hugging Face Spaces
+if __name__ == "__main__":
+    import uvicorn
+
+    uvicorn.run(app, host="0.0.0.0", port=7860)