🎙️ AI Voice Button

🔘 Hardware: Buttons zum Bestellen

⚡ Tech Stack

🔄 Ablauf (Workflow)

💻 Code-Grundgerüst

# voice_button.py
import keyboard
import sounddevice as sd
import numpy as np
import wavio
import openai
import requests
from elevenlabs import generate, stream, set_api_key
import os
from datetime import datetime

# API Keys
openai.api_key = os.getenv("OPENAI_API_KEY")
set_api_key(os.getenv("ELEVENLABS_API_KEY"))

# Config
SAMPLE_RATE = 16000
RECORDING = False
audio_buffer = []

def record_callback(indata, frames, time, status):
    global audio_buffer
    if RECORDING:
        audio_buffer.append(indata.copy())

def start_recording():
    global RECORDING, audio_buffer
    print("🔴 Aufnahme startet...")
    RECORDING = True
    audio_buffer = []
    # LED einschalten (falls vorhanden)

def stop_recording():
    global RECORDING
    RECORDING = False
    print("⏹️ Aufnahme stoppt...")
    
    # Audio speichern
    if audio_buffer:
        audio = np.concatenate(audio_buffer, axis=0)
        filename = f"recording_{datetime.now().strftime('%Y%m%d_%H%M%S')}.wav"
        wavio.write(filename, audio, SAMPLE_RATE, sampwidth=2)
        
        # Verarbeitung
        process_audio(filename)

def process_audio(filename):
    # 1. Transkription
    with open(filename, "rb") as audio_file:
        transcript = openai.audio.transcriptions.create(
            model="whisper-1",
            file=audio_file
        )
    print(f"📝 Gesagt: {transcript.text}")
    
    # 2. LLM Antwort
    response = openai.chat.completions.create(
        model="gpt-4",
        messages=[
            {"role": "system", "content": "Du bist ein freundlicher Ausstellungs-Guide. Antworte kurz und prägnant auf Deutsch."},
            {"role": "user", "content": transcript.text}
        ]
    )
    answer = response.choices[0].message.content
    print(f"🤖 Antwort: {answer}")
    
    # 3. ElevenLabs TTS (Streaming)
    audio_stream = generate(
        text=answer,
        voice="Scarlett",  # oder: "Adam", "Bella", "Antoni"
        model="eleven_turbo_v2",
        stream=True
    )
    stream(audio_stream)
    print("✅ Fertig!")

# Keyboard Listener
def on_space_press(e):
    if e.event_type == 'down' and not RECORDING:
        start_recording()
    elif e.event_type == 'up' and RECORDING:
        stop_recording()

print("🎙️ Voice Button bereit. Drück Space...")
keyboard.hook_key('space', on_space_press)
keyboard.wait()

💰 Kosten-Schätzung

📦 Installation (Mac Mini)

# 1. Python Dependencies
pip install openai elevenlabs sounddevice wavio numpy keyboard

# 2. ElevenLabs API Key holen
# → https://elevenlabs.io/app/settings/api-keys

# 3. OpenAI API Key holen  
# → https://platform.openai.com/api-keys

# 4. Environment Variables setzen
export OPENAI_API_KEY="sk-..."
export ELEVENLABS_API_KEY="..."

# 5. App starten
python voice_button.py

🚀 Alternative: Realtime Voice (Pro-Level)

Für echten "Live-Gespräch"-Modus statt Knopf-Drücken:

• OpenAI Realtime API: WebRTC-basiert, echter Dialog
• ElevenLabs Conv. AI: Eigene Voice-API mit Unterbrechungen

Das ist komplexer, aber fühlt sich an wie ChatGPT Voice.

📋 Shopping-Checkliste

📦 Bestellung – 60mm USB Button