init: Kira — AI body double with Honcho memory

Full voice pipeline (Whisper STT -> DeepSeek LLM -> OpenAI TTS),
animated SVG avatar (Live2D-ready), girly-pop UI, lofi music,
timer/notes/pets/wardrobe widgets, 10 background scenes with
particle effects, Honcho cross-session memory.

backend/services/stt.py

@@ -0,0 +1,27 @@
+"""Speech-to-text via OpenAI Whisper API"""
+
+import logging
+from openai import AsyncOpenAI
+from config import settings
+
+logger = logging.getLogger("kira.stt")
+
+
+def _get_client() -> AsyncOpenAI:
+    return AsyncOpenAI(api_key=settings.openai_api_key)
+
+
+async def transcribe_audio(audio_bytes: bytes) -> str | None:
+    """Transcribe audio bytes to text using Whisper API."""
+    try:
+        client = _get_client()
+        transcript = await client.audio.transcriptions.create(
+            model="whisper-1",
+            file=("audio.webm", audio_bytes, "audio/webm"),
+            language="en",
+            response_format="text",
+        )
+        return transcript.strip() if transcript and transcript.strip() else None
+    except Exception as e:
+        logger.error(f"STT error: {e}")
+        return None