# Whisperのインストール
pip3 install openai-whisper

# 音声入力用ライブラリ
pip3 install pyaudio numpy torch

import whisper
import pyaudio
import numpy as np
import threading
import queue
import time
import torch

class WhisperTranscriber:
    def __init__(self, model_size='base', language='japanese'):
        """
        Whisper音声認識クラス
        """
        self.model = whisper.load_model(model_size)

        for param in self.model.parameters():
            print(f"モデルのデバイス: {param.device}")
            break  # 最初のパラメータだけ確認すれば十分
        
        # 音声入力設定
        self.pyaudio = pyaudio.PyAudio()
        self.sample_rate = 16000
        self.chunk_duration = 3
        self.chunks_per_inference = int(self.sample_rate * self.chunk_duration)
        
        # スレッディング用キュー
        self.audio_queue = queue.Queue()
        self.stop_event = threading.Event()
        
        self.language = language

    def transcribe_audio(self):
        """
        リアルタイム音声文字起こし
        """
        if self.model is None:
            print("モデルの初期化に失敗しました")
            return

        audio_data = []
        while not self.stop_event.is_set():
            try:
                chunk = self.audio_queue.get(timeout=1)
                audio_data.extend(chunk)

                if len(audio_data) >= self.chunks_per_inference:
                    # テンソル変換
                    audio_np = np.array(audio_data[:self.chunks_per_inference])
                    
                    # 推論を実行
                    result = self.model.transcribe(
                        audio_np, 
                        language=self.language,
                        fp16=False
                    )
                    
                    print(f"文字起こし結果: {result['text']}")
                    audio_data = audio_data[self.chunks_per_inference:]

            except queue.Empty:
                continue
            except Exception as e:
                print(f"推論中にエラー: {e}")

    def audio_callback(self, in_data, frame_count, time_info, status):
        """
        音声データ処理用コールバック
        """
        audio_chunk = np.frombuffer(in_data, dtype=np.float32)
        self.audio_queue.put(audio_chunk)
        return (None, pyaudio.paContinue)

    def start_recording(self):
        """
        音声録音開始
        """
        self.stream = self.pyaudio.open(
            format=pyaudio.paFloat32,
            channels=1,
            rate=self.sample_rate,
            input=True,
            frames_per_buffer=self.chunks_per_inference,
            stream_callback=self.audio_callback
        )
        print("音声録音を開始")

    def run(self):
        """
        音声認識プロセス実行
        """
        self.start_recording()
        transcribe_thread = threading.Thread(target=self.transcribe_audio)
        transcribe_thread.start()

        try:
            transcribe_thread.join()
        except KeyboardInterrupt:
            print("\n音声認識を終了")
            self.stop_event.set()
            self.stream.stop_stream()
            self.stream.close()
            self.pyaudio.terminate()

def main():
    transcriber = WhisperTranscriber(
        model_size='base',  
        language='japanese'
    )
    transcriber.run()

if __name__ == "__main__":
    main()

import torch
print(torch.mps.is_available())