first tests

scripts/tools_test.py

@@ -0,0 +1,64 @@
+"""Exercise "Voice + tools" over the WebSocket: typed requests, then a spoken one (made with the server's TTS)."""
+
+import asyncio
+import json
+
+import numpy as np
+import websockets
+
+URL = "ws://127.0.0.1:8000/ws"
+PROMPTS = [
+    "What's the weather like in Paris right now?",
+    "What is 17 times 23, divided by 4?",
+    "Please make the screen green.",
+    "What time is it in Tokyo?",
+    "Tell me a fun fact about octopuses.",  # should answer without a tool
+]
+
+
+async def turn(ws, send):
+    await send()
+    pcm, events = [], []
+    while True:
+        m = await ws.recv()
+        if isinstance(m, bytes):
+            pcm.append(np.frombuffer(m, dtype=np.float32))
+            continue
+        m = json.loads(m)
+        if m["type"] in ("tool_call", "tool_result", "ui"):
+            events.append({k: v for k, v in m.items() if k not in ("t_ms",)})
+        if m["type"] in ("done", "error"):
+            return m, events, np.concatenate(pcm) if pcm else np.zeros(0, np.float32)
+
+
+async def main():
+    async with websockets.connect(URL, max_size=None) as ws:
+        await ws.recv()
+        await ws.send(json.dumps({"type": "settings", "mode": "tools", "keep_history": False}))
+        for p in PROMPTS:
+            done, events, pcm = await turn(ws, lambda: ws.send(json.dumps({"type": "text", "text": p})))
+            print(f"\n> {p}")
+            for e in events:
+                print("  ", e)
+            s = done.get("stats", {})
+            print(f"  reply: {done.get('text')!r}\n  rounds={s.get('rounds')} audio={s.get('audio_out_s')}s "
+                  f"first_text={s.get('first_text_ms')}ms gen={s.get('gen_ms')}ms")
+
+        # spoken request: synthesize it with TTS, then send it as mic audio
+        q = "What's the weather in Stockholm?"
+        _, _, speech = await turn(ws, lambda: ws.send(json.dumps({"type": "tts", "text": q})))
+
+        async def speak():
+            await ws.send(json.dumps({"type": "start", "rate": 24000}))
+            for i in range(0, len(speech), 2048):
+                await ws.send(speech[i:i + 2048].tobytes())
+            await ws.send(json.dumps({"type": "stop"}))
+
+        done, events, _ = await turn(ws, speak)
+        print(f"\n> (spoken) {q}")
+        for e in events:
+            print("  ", e)
+        print(f"  reply: {done.get('text')!r} rounds={done['stats'].get('rounds')}")
+
+
+asyncio.run(main())