maciej-bot-2.0/voice_bot_live.py

import discord
from discord.ext import tasks, commands
from faster_whisper import WhisperModel
import os
import asyncio
import wave
import time
import ctypes.util

# ============================
# 0. FIX NA OPUS (DLA PEWNOŚCI)
# ============================
# Próba załadowania biblioteki systemowej, żeby uciszyć część błędów
opus_path = ctypes.util.find_library('opus')
if opus_path:
    try:
        discord.opus.load_opus(opus_path)
    except:
        pass

# ============================
# 1. KONFIGURACJA MODELU
# ============================
print("⏳ Ładowanie Faster-Whisper...")
model = WhisperModel("base", device="cpu", compute_type="int8")
print("✅ Model gotowy!")

# ============================
# 2. KLASA ODBIORNIKA
# ============================
class LiveSink(discord.sinks.Sink):
    def __init__(self):
        super().__init__()
        self.user_buffers = {}
        self.last_process_time = {}

    # Fix dla biblioteki py-cord (obsługa startu)
    def init(self, vc): 
        self.vc = vc
        try:
            super().init(vc)
        except TypeError:
            pass

    # Fix dla biblioteki py-cord (obsługa zapisu danych)
    @discord.sinks.Filters.container
    def write(self, data, user): 
        if user not in self.user_buffers:
            self.user_buffers[user] = bytearray()
            self.last_process_time[user] = time.time()
        
        self.user_buffers[user] += data

    def get_audio_chunk(self, user):
        if user in self.user_buffers:
            data = self.user_buffers[user]
            self.user_buffers[user] = bytearray() # Reset bufora
            self.last_process_time[user] = time.time()
            return data
        return None

# ============================
# 3. LOGIKA PRZETWARZANIA
# ============================
intents = discord.Intents.default()
intents.message_content = True
bot = commands.Bot(command_prefix="!", intents=intents)

current_sink = None

async def transcode_and_transcribe(user_id, pcm_data):
    # Unikalna nazwa pliku
    temp_filename = f"live_{user_id}_{int(time.time()*1000)}.wav"
    
    try:
        # Zapis PCM do WAV (format wymagany przez Whisper)
        with wave.open(temp_filename, 'wb') as wav_file:
            wav_file.setnchannels(2)
            wav_file.setsampwidth(2)
            wav_file.setframerate(48000)
            wav_file.writeframes(pcm_data)

        # Uruchomienie modelu w tle
        loop = asyncio.get_event_loop()
        def run_whisper():
            segments, _ = model.transcribe(temp_filename, language="pl", beam_size=1)
            return " ".join([s.text for s in segments])

        text = await loop.run_in_executor(None, run_whisper)
        
        # Filtrujemy puste wiadomości i halucynacje "dzięki"
        if text.strip() and "dzięki" not in text.lower():
            user = await bot.fetch_user(user_id)
            print(f"🔴 LIVE {user.name}: {text}")

    except Exception as e:
        print(f"Błąd transkrypcji: {e}")
    finally:
        if os.path.exists(temp_filename):
            os.remove(temp_filename)

@tasks.loop(seconds=3.0)
async def live_transcription_loop():
    if current_sink is None:
        return

    try:
        # Iterujemy po liście userów
        user_ids = list(current_sink.user_buffers.keys())
        for user_id in user_ids:
            # Sprawdzamy czy user coś mówił (czy bufor > 100KB)
            if len(current_sink.user_buffers[user_id]) > 100000: 
                pcm_data = current_sink.get_audio_chunk(user_id)
                if pcm_data:
                    asyncio.create_task(transcode_and_transcribe(user_id, pcm_data))
    except Exception:
        pass

# --- TO JEST NOWA FUNKCJA NAPRAWIAJĄCA BŁĄD ---
async def dummy_callback(sink, *args):
    # Ta funkcja musi być async, inaczej bot wywala błąd "coroutine required"
    return

# ============================
# 4. KOMENDY
# ============================
@bot.event
async def on_ready():
    print(f'🚀 Bot Live gotowy: {bot.user}')

@bot.command()
async def join(ctx):
    if ctx.author.voice:
        await ctx.author.voice.channel.connect()
        await ctx.send("Dołączono.")
    else:
        await ctx.send("Wejdź na kanał.")

@bot.command()
async def start(ctx):
    global current_sink
    if not ctx.voice_client:
        return await ctx.send("Nie jestem połączony.")

    print("Rozpoczynam LIVE transkrypcję...")
    current_sink = LiveSink()
    
    # Tu była lambda, która psuła bota. Teraz jest dummy_callback.
    ctx.voice_client.start_recording(
        current_sink,
        dummy_callback, 
        ctx.channel
    )
    
    live_transcription_loop.start()
    await ctx.send("Nasłuchuję w trybie LIVE...")

@bot.command()
async def stop(ctx):
    global current_sink
    if ctx.voice_client:
        ctx.voice_client.stop_recording() # To wywoła dummy_callback
        live_transcription_loop.stop()
        current_sink = None
        await ctx.send("Zatrzymano.")

token = os.getenv("DISCORD_TOKEN")
if token:
    bot.run(token)
else:
    print("Brak tokena!")