konan-ai-assistant/konan_client.py

"""
Konan Voice Client — PC Windows
Push-to-talk : maintiens ESPACE pour parler, relâche pour envoyer.
Ctrl+C pour quitter.

Installation :
    pip install sounddevice scipy openai-whisper requests keyboard pyttsx3
"""

import sounddevice as sd
import scipy.io.wavfile as wav
import whisper
import requests
import pyttsx3
import keyboard
import numpy as np
import tempfile
import os
import sys

KONAN_API = "http://192.168.1.21:8765/api/chat"
SAMPLE_RATE = 16000
USER_ID = "pc_user"

print("Chargement de Whisper tiny...")
model = whisper.load_model("tiny")
print("Whisper prêt.")

engine = pyttsx3.init()
engine.setProperty("rate", 165)
# Voix française si disponible
voices = engine.getProperty("voices")
for v in voices:
    if "fr" in v.id.lower() or "french" in v.name.lower():
        engine.setProperty("voice", v.id)
        break

def speak(text):
    print(f"Konan : {text}")
    engine.say(text)
    engine.runAndWait()

def record_until_release():
    print("🎙️  Enregistrement... (relâche Espace pour envoyer)")
    chunks = []
    stream = sd.InputStream(samplerate=SAMPLE_RATE, channels=1, dtype="int16")
    stream.start()
    while keyboard.is_pressed("space"):
        chunk, _ = stream.read(1024)
        chunks.append(chunk)
    stream.stop()
    stream.close()
    if not chunks:
        return None
    audio = np.concatenate(chunks, axis=0).flatten()
    return audio

def transcribe(audio):
    with tempfile.NamedTemporaryFile(suffix=".wav", delete=False) as f:
        tmp_path = f.name
        wav.write(tmp_path, SAMPLE_RATE, audio)
    result = model.transcribe(tmp_path, language="fr")
    try:
        os.unlink(tmp_path)
    except Exception:
        pass
    return result["text"].strip()

def ask_konan(text):
    try:
        r = requests.post(KONAN_API, json={"text": text, "user_id": USER_ID}, timeout=120)
        data = r.json()
        return data.get("response", "Pas de réponse.")
    except Exception as e:
        return f"Erreur connexion : {e}"

speak("Konan est prêt. Maintiens Espace pour parler.")
print("\n▶  Maintiens ESPACE pour parler. Ctrl+C pour quitter.\n")

try:
    while True:
        keyboard.wait("space")
        audio = record_until_release()
        if audio is None or len(audio) < SAMPLE_RATE * 0.3:
            print("(trop court, ignoré)")
            continue
        print("Transcription...")
        text = transcribe(audio)
        if not text:
            print("(rien détecté)")
            continue
        print(f"Toi : {text}")
        print("Konan réfléchit...")
        response = ask_konan(text)
        speak(response)
except KeyboardInterrupt:
    print("\nAu revoir.")