【ReazonSpeech】日本語音声の文字起こしにチャレンジ（3回目）。「reazonspeech-k2-v2」というモデルを試してみた。

はじめに

以前にもReazonSpeechを試しています。
touch-sp.hatenablog.com
touch-sp.hatenablog.com
今回3回目です。

環境

Windows 11
CUDA 12.6
Python 3.13.3

PySide6でGUIを作っています。

torch==2.7.0+cu126
PySide6==6.9.0

環境構築

pip install torch==2.7.0+cu126 --index-url https://download.pytorch.org/whl/cu126
pip install pyside6
git clone https://github.com/reazon-research/ReazonSpeech
pip install ReazonSpeech/pkg/k2-asr

Pythonスクリプト

import sys
import os
from pathlib import Path
from datetime import datetime
from queue import Queue
from threading import Thread
from PySide6.QtWidgets import QApplication, QMainWindow, QPushButton, QVBoxLayout, QWidget, QTextEdit, QLabel
from PySide6.QtCore import QUrl, Signal, QObject, QTimer
from PySide6.QtMultimedia import QMediaRecorder, QMediaCaptureSession, QAudioInput, QMediaFormat
from reazonspeech.k2.asr import load_model, transcribe, audio_from_path

model = load_model()

class TranscriptionSignals(QObject):
    transcription_done = Signal(str)
    error_occurred = Signal(str)

class AudioTranscriptionApp(QMainWindow):
    def __init__(self):
        super().__init__()
        self.setWindowTitle("音声録音と自動文字起こし")
        self.setGeometry(100, 100, 400, 300)

        layout = QVBoxLayout()

        self.record_button = QPushButton("録音開始")
        self.record_button.clicked.connect(self.toggle_recording)
        layout.addWidget(self.record_button)

        self.status_label = QLabel("待機中")
        layout.addWidget(self.status_label)

        self.text_edit = QTextEdit()
        layout.addWidget(self.text_edit)

        container = QWidget()
        container.setLayout(layout)
        self.setCentralWidget(container)

        # メディアコンポーネントの設定
        self.audio_input = QAudioInput()
        self.media_recorder = QMediaRecorder()
        self.capture_session = QMediaCaptureSession()
        self.capture_session.setAudioInput(self.audio_input)
        self.capture_session.setRecorder(self.media_recorder)

        # 録音フォーマットの設定
        audio_settings = self.media_recorder.mediaFormat()
        audio_settings.setFileFormat(QMediaFormat.FileFormat.Wave)
        audio_settings.setAudioCodec(QMediaFormat.AudioCodec.Wave)
        self.media_recorder.setMediaFormat(audio_settings)

        # 品質設定
        self.media_recorder.setQuality(QMediaRecorder.Quality.HighQuality)

        self.is_recording = False
        self.temp_file = None

        # 録音キューと文字起こしスレッドの設定
        self.audio_queue = Queue()
        self.transcription_thread = Thread(target=self.transcribe_audio_thread, daemon=True)
        self.transcription_thread.start()

        # シグナルの設定
        self.signals = TranscriptionSignals()
        self.signals.transcription_done.connect(self.update_transcription)
        self.signals.error_occurred.connect(self.show_error)

        # メディアレコーダーのシグナル接続
        self.media_recorder.recorderStateChanged.connect(self.on_recorder_state_changed)

    def toggle_recording(self):
        if not self.is_recording:
            self.start_recording()
        else:
            self.stop_recording()

    def start_recording(self):
        now = datetime.now()
        current_time = now.strftime("%Y%m%d_%H%M%S")
        current_directory = Path.cwd()
        self.temp_file = Path(current_directory, f"{current_time}.wav")

        self.media_recorder.setOutputLocation(QUrl.fromLocalFile(str(self.temp_file)))
        self.media_recorder.record()
        self.is_recording = True
        self.record_button.setText("録音停止")
        self.status_label.setText("録音中...")

    def stop_recording(self):
        self.media_recorder.stop()
        self.is_recording = False
        self.record_button.setText("録音開始")
        self.status_label.setText("録音停止中...")

    def on_recorder_state_changed(self, state):
        if state == QMediaRecorder.StoppedState:
            QTimer.singleShot(1000, self.process_recorded_file)

    def process_recorded_file(self):
        if self.temp_file and os.path.exists(self.temp_file):
            self.status_label.setText("文字起こし中...")
            self.audio_queue.put(str(self.temp_file))
            self.temp_file = None
        else:
            self.signals.error_occurred.emit("録音ファイルが見つかりません。")

    def transcribe_audio_thread(self):
        while True:
            audio_file = self.audio_queue.get()
            try:
                audio = audio_from_path(audio_file)
                ret = transcribe(model, audio)
                self.signals.transcription_done.emit(ret.text)
            except Exception as e:
                self.signals.error_occurred.emit(f"文字起こし中にエラーが発生しました: {str(e)}")
            finally:
                self.audio_queue.task_done()

    def update_transcription(self, text):
        self.text_edit.append(text)
        self.status_label.setText("待機中")

    def show_error(self, error_message):
        self.status_label.setText("エラー")
        self.text_edit.append(f"エラー: {error_message}")

if __name__ == "__main__":
    app = QApplication(sys.argv)
    window = AudioTranscriptionApp()
    window.show()
    sys.exit(app.exec())

このスクリプトはこちらとほとんど同じです。
touch-sp.hatenablog.com