From b209394d8df66edcd8436192f34b1304371b7baf Mon Sep 17 00:00:00 2001
From: Till-Immer <till@immers.ch>
Date: Sun, 5 Jul 2026 16:17:35 +0200
Subject: [PATCH] added voice pipiline

---
 .gitignore                         |   4 +
 __pycache__/config.cpython-311.pyc | Bin 360 -> 602 bytes
 voice_nora.py                      | 132 +++++++++++++++++++++++++++++
 3 files changed, 136 insertions(+)
 create mode 100644 .gitignore
 create mode 100644 voice_nora.py

diff --git a/.gitignore b/.gitignore
new file mode 100644
index 0000000..b7d4344
--- /dev/null
+++ b/.gitignore
@@ -0,0 +1,4 @@
+__pycache__/
+.env
+venv/
+
diff --git a/__pycache__/config.cpython-311.pyc b/__pycache__/config.cpython-311.pyc
index 65f72c1b02c0478f0f6daecc2ddf50ed38ed1d59..766f5f60727d9bd2a28a515f7063d741dbb84c34 100644
GIT binary patch
delta 386
zcmaFCbc=;=IWI340}$*|^vV)voX97^cy6M)Z#@%3Dnk}CNCE^>Se7v|Fsue*2#8`u
zlV?SfXG4=`Lz8DmlV?Yl=SX1;X3*rASSRMs<rp3uALQ!p>F;+-5X5$Lc6JR8j`wzr
zjQ4c8B?=M>c6APN4MCE*#p~qi66Ejf9q;S!;_7pY-PbuFJ~YTjlXK#BIguh(pg)S(
zKm<F8-~bX6AL?`Q18I;ai#sMuG3qDsigg6_L{6x=$g6UNSEYgLhKTG0rzxHn<P0y0
z7+n!Dy1*cKLqc%|(;UtVN|qNTtgc8{U0@J>ARsw`X$t2I{fh!xR|K>gcpq>HcPRB}
VPVl_QC3l5Ou7Mo{i+F+d004s=RX_j$

delta 163
zcmcb`@`8zPIWI340}$*z>5<jWFp*D!F=(Q?uX74Z6l)4&FoP!ROOULeCgUycoczR;
z_>}yT)V#7=eD1Eko_?P3jsc$W-mZ~V0_mx_nR%JIMtY{YX*r3-8JcX9;~C|6ZgKhg
zySVzq`#Ji$PM*l9&&2~YjS+~8-6o%6)Rz~Onjkbq@`AMXMM0e_f;tU+4|wGpxInN7
G<SYQ5Vl1lw

diff --git a/voice_nora.py b/voice_nora.py
new file mode 100644
index 0000000..fb63e17
--- /dev/null
+++ b/voice_nora.py
@@ -0,0 +1,132 @@
+import asyncio
+import os
+import numpy as np
+import pyaudio
+from openwakeword.model import Model
+from faster_whisper import WhisperModel
+from edge_tts import Communicate
+import wave
+import webrtcvad  # Für bessere Sprach-Erkennung
+
+from agent import Agent
+from util.logger import get_logger
+
+logger = get_logger("VOICE")
+
+
+class VoiceNora:
+    def __init__(self):
+        self.agent = Agent()
+        self.running = True
+        self.vad = webrtcvad.Vad(2)  # Aggressivität 0-3
+
+        # Wake Word
+        self.ww_model = Model(wakeword="nora")
+
+        # STT
+        self.stt_model = WhisperModel("tiny", device="cpu", compute_type="int8")
+
+        print("\n🎤 Voice N.O.R.A gestartet")
+        print("   Wakeword: 'Nora'")
+        print("   Weibliche Stimme aktiviert\n")
+
+    async def listen_for_wakeword(self):
+        pa = pyaudio.PyAudio()
+        stream = pa.open(format=pyaudio.paInt16, channels=1, rate=16000,
+                         input=True, frames_per_buffer=1024)
+
+        print("⏳ Lausche auf 'Nora'...")
+
+        while self.running:
+            try:
+                audio_chunk = stream.read(1024, exception_on_overflow=False)
+                audio_np = np.frombuffer(audio_chunk, dtype=np.int16)
+
+                if self.ww_model.predict(audio_np).get("nora", 0) > 0.58:
+                    print("\n✅ 'Nora' erkannt!")
+                    await self.process_voice_command()
+            except:
+                pass
+
+        stream.stop_stream()
+        stream.close()
+        pa.terminate()
+
+    async def process_voice_command(self):
+        print("🎤 Ich höre zu... (sprich natürlich)")
+
+        audio_data = await self.record_with_vad(max_duration=12)
+
+        if len(audio_data) < 8000:  # zu kurz
+            print("❌ Zu kurz.")
+            return
+
+        # STT
+        segments, _ = self.stt_model.transcribe(
+            audio_data, language="de", beam_size=5, vad_filter=True
+        )
+        text = " ".join(segment.text for segment in segments).strip()
+
+        if not text:
+            print("❌ Konnte nichts verstehen.")
+            return
+
+        print(f"👤 Du: {text}")
+
+        response = await self.agent.run(text)
+        print(f"🗣️ N.O.R.A: {response}")
+
+        await self.speak(response)
+
+    async def record_with_vad(self, max_duration=12):
+        """Aufnahme mit Sprach-Erkennung (endet wenn du aufhörst zu sprechen)"""
+        pa = pyaudio.PyAudio()
+        stream = pa.open(format=pyaudio.paInt16, channels=1, rate=16000,
+                         input=True, frames_per_buffer=320)
+
+        frames = []
+        silence_count = 0
+        max_silence = 25  # ca. 1,5 Sekunden Stille
+
+        for _ in range(int(16000 / 320 * max_duration)):
+            chunk = stream.read(320, exception_on_overflow=False)
+            frames.append(chunk)
+
+            # VAD prüfen
+            is_speech = self.vad.is_speech(chunk, 16000)
+            if not is_speech:
+                silence_count += 1
+            else:
+                silence_count = 0
+
+            if silence_count > max_silence and len(frames) > 30:
+                break  # aufhören wenn lange Stille
+
+        stream.stop_stream()
+        stream.close()
+        pa.terminate()
+
+        return b''.join(frames)
+
+    async def speak(self, text: str):
+        """Edge TTS - weibliche Stimme"""
+        try:
+            communicate = Communicate(text, voice="de-DE-AmalaNeural")  # Weiblich, natürlich
+
+            await communicate.save("response.mp3")
+            os.system("mpg123 -q response.mp3")
+
+        except Exception as e:
+            logger.error(f"TTS Fehler: {e}")
+
+async def main():
+    voice = VoiceNora()
+    try:
+        await voice.listen_for_wakeword()
+    except KeyboardInterrupt:
+        print("\n\n👋 N.O.R.A wird beendet.")
+    except Exception as e:
+        logger.error(str(e))
+
+if __name__ == "__main__":
+    asyncio.run(main())