Add "Use GPU" checkbox to Speech to Text

It defaults to on, and the job will automatically retry without GPU if it fails. If the --no-gpu job succeeds after retrying, it changes the checkbox default value in settings.

Author: ddennedy

src/dialogs/transcribeaudiodialog.cpp

@@ -138,9 +138,14 @@ TranscribeAudioDialog::TranscribeAudioDialog(const QString &trackName, QWidget *
     grid->addWidget(m_nonspoken, 4, 0, Qt::AlignRight);
     grid->addWidget(new QLabel(tr("Include non-spoken sounds")), 4, 1, Qt::AlignLeft);
 
+    m_useGpu = new QCheckBox(this);
+    m_useGpu->setCheckState(Settings.whisperUseGpu() ? Qt::Checked : Qt::Unchecked);
+    grid->addWidget(m_useGpu, 5, 0, Qt::AlignRight);
+    grid->addWidget(new QLabel(tr("Use GPU")), 5, 1, Qt::AlignLeft);
+
     QLabel *tracksLabel = new QLabel(tr("Tracks with speech"));
     tracksLabel->setToolTip(tr("Select tracks that contain speech to be transcribed."));
-    grid->addWidget(tracksLabel, 5, 0, Qt::AlignRight);
+    grid->addWidget(tracksLabel, 6, 0, Qt::AlignRight);
     m_trackList = new QListWidget(this);
     m_trackList->setSelectionMode(QAbstractItemView::NoSelection);
     m_trackList->setSizeAdjustPolicy(QAbstractScrollArea::AdjustToContentsOnFirstShow);
@@ -172,7 +177,7 @@ TranscribeAudioDialog::TranscribeAudioDialog(const QString &trackName, QWidget *
             }
         }
     }
-    grid->addWidget(m_trackList, 5, 1, Qt::AlignLeft);
+    grid->addWidget(m_trackList, 6, 1, Qt::AlignLeft);
 
     // The config section is a single widget with a unique grid layout inside of it.
     // The config section is hidden by hiding the config widget (and the layout it contains)
@@ -274,7 +279,7 @@ TranscribeAudioDialog::TranscribeAudioDialog(const QString &trackName, QWidget *
 
     configLayout->addWidget(m_table, 4, 0, 1, 3);
 
-    grid->addWidget(m_configWidget, 6, 0, 1, 2);
+    grid->addWidget(m_configWidget, 7, 0, 1, 2);
 
     // Add a button box to the dialog
     m_buttonBox = new QDialogButtonBox(QDialogButtonBox::Ok | QDialogButtonBox::Cancel);
@@ -294,7 +299,7 @@ TranscribeAudioDialog::TranscribeAudioDialog(const QString &trackName, QWidget *
         m_configWidget->setVisible(false);
     }
     m_buttonBox->addButton(configButton, QDialogButtonBox::ActionRole);
-    grid->addWidget(m_buttonBox, 7, 0, 1, 2);
+    grid->addWidget(m_buttonBox, 8, 0, 1, 2);
     connect(m_buttonBox,
             SIGNAL(clicked(QAbstractButton *)),
             this,
@@ -321,6 +326,7 @@ void TranscribeAudioDialog::onButtonClicked(QAbstractButton *button)
 {
     QDialogButtonBox::ButtonRole role = m_buttonBox->buttonRole(button);
     if (role == QDialogButtonBox::AcceptRole) {
+        Settings.setWhisperUseGpu(m_useGpu->checkState() == Qt::Checked);
         LOG_DEBUG() << "Accept";
         accept();
     } else if (role == QDialogButtonBox::RejectRole) {
@@ -378,6 +384,11 @@ bool TranscribeAudioDialog::includeNonspoken()
     return m_nonspoken->checkState() == Qt::Checked;
 }
 
+bool TranscribeAudioDialog::useGpu()
+{
+    return m_useGpu->checkState() == Qt::Checked;
+}
+
 void TranscribeAudioDialog::showEvent(QShowEvent *event)
 {
     QDialog::showEvent(event);

src/dialogs/transcribeaudiodialog.h

@@ -43,6 +43,7 @@ class TranscribeAudioDialog : public QDialog
     bool translate();
     int maxLineLength();
     bool includeNonspoken();
+    bool useGpu();
 
 protected:
     virtual void showEvent(QShowEvent *event) override;
@@ -63,6 +64,7 @@ private slots:
     QCheckBox *m_translate;
     QSpinBox *m_maxLength;
     QCheckBox *m_nonspoken;
+    QCheckBox *m_useGpu;
     QListWidget *m_trackList;
     QWidget *m_configWidget;
     QLineEdit *m_exeLabel;

src/docks/subtitlesdock.cpp

@@ -1341,7 +1341,8 @@ void SubtitlesDock::speechToText()
                                             tmpSrt->fileName(),
                                             dialog.language(),
                                             dialog.translate(),
-                                            dialog.maxLineLength());
+                                            dialog.maxLineLength(),
+                                            dialog.useGpu());
     // Ensure the language code is 3 character (part 2)
     QString langCode = dialog.language();
     QLocale::Language lang = QLocale::codeToLanguage(langCode);

src/jobs/whisperjob.cpp

@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2024 Meltytech, LLC
+ * Copyright (c) 2024-2026 Meltytech, LLC
  *
  * This program is free software: you can redistribute it and/or modify
  * it under the terms of the GNU General Public License as published by
@@ -33,13 +33,16 @@ WhisperJob::WhisperJob(const QString &name,
                        const QString &lang,
                        bool translate,
                        int maxLength,
+                       bool useGpu,
                        QThread::Priority priority)
     : AbstractJob(name, priority)
     , m_iWavFile(iWavFile)
     , m_oSrtFile(oSrtFile)
     , m_lang(lang)
     , m_translate(translate)
     , m_maxLength(maxLength)
+    , m_useGpu(useGpu)
+    , m_retryingWithoutGpu(false)
     , m_previousPercent(0)
 {
     setTarget(oSrtFile);
@@ -60,25 +63,29 @@ void WhisperJob::start()
     QString of = m_oSrtFile;
     of.remove(".srt");
     QStringList args;
-    args << "-f" << m_iWavFile;
-    args << "-m" << modelPath;
-    args << "-l" << m_lang;
+    args << "--file" << m_iWavFile;
+    args << "--model" << modelPath;
+    args << "--language" << m_lang;
     if (m_translate) {
-        args << "-tr";
+        args << "--translate";
     }
-    args << "-of" << of;
-    args << "-osrt";
-    args << "-pp";
-    args << "-ml" << QString::number(m_maxLength);
-    args << "-sow";
+    args << "--output-file" << of;
+    args << "--output-srt";
+    args << "--print-progress";
+    args << "--max-len" << QString::number(m_maxLength);
+    args << "--split-on-word";
 
 #if QT_POINTER_SIZE == 4
     // Limit to 1 rendering thread on 32-bit process to reduce memory usage.
     auto threadCount = 1;
 #else
     auto threadCount = qMax(1, QThread::idealThreadCount() - 1);
 #endif
-    args << "-t" << QString::number(threadCount);
+    args << "--threads" << QString::number(threadCount);
+
+    if (!m_useGpu || m_retryingWithoutGpu) {
+        args << "--no-gpu";
+    }
 
     LOG_DEBUG() << whisperPath + " " + args.join(' ');
     AbstractJob::start(whisperPath, args);
@@ -116,3 +123,21 @@ void WhisperJob::onReadyRead()
         }
     } while (!msg.isEmpty());
 }
+
+void WhisperJob::onFinished(int exitCode, QProcess::ExitStatus exitStatus)
+{
+    if ((exitStatus != QProcess::NormalExit || exitCode != 0) && !stopped() && m_useGpu
+        && !m_retryingWithoutGpu) {
+        m_retryingWithoutGpu = true;
+        m_previousPercent = 0;
+        QString message(tr("Speech to Text job failed; trying again without GPU."));
+        MAIN.showStatusMessage(message);
+        appendToLog(message.append("\n"));
+        QTimer::singleShot(0, this, &WhisperJob::start);
+        return;
+    }
+    if (m_retryingWithoutGpu && exitStatus == QProcess::NormalExit && exitCode == 0) {
+        Settings.setWhisperUseGpu(false);
+    }
+    AbstractJob::onFinished(exitCode, exitStatus);
+}

src/jobs/whisperjob.h

@@ -32,6 +32,7 @@ class WhisperJob : public AbstractJob
                const QString &lang,
                bool translate,
                int maxLength,
+               bool useGpu = true,
                QThread::Priority priority = Settings.jobPriority());
     virtual ~WhisperJob();
 
@@ -41,13 +42,16 @@ public slots:
 
 protected slots:
     void onReadyRead();
+    void onFinished(int exitCode, QProcess::ExitStatus exitStatus) override;
 
 private:
     const QString m_iWavFile;
     const QString m_oSrtFile;
     const QString m_lang;
     const bool m_translate;
     const int m_maxLength;
+    const bool m_useGpu;
+    bool m_retryingWithoutGpu;
     int m_previousPercent;
 };
 

src/settings.cpp

@@ -1699,6 +1699,16 @@ QString ShotcutSettings::whisperModel()
     return settings.value("subtitles/whisperModel", "").toString();
 }
 
+void ShotcutSettings::setWhisperUseGpu(bool b)
+{
+    settings.setValue("subtitles/whisperUseGpu", b);
+}
+
+bool ShotcutSettings::whisperUseGpu() const
+{
+    return settings.value("subtitles/whisperUseGpu", true).toBool();
+}
+
 void ShotcutSettings::setNotesZoom(int zoom)
 {
     settings.setValue("notes/zoom", zoom);

src/settings.h

@@ -370,6 +370,8 @@ class ShotcutSettings : public QObject
     QString whisperExe();
     void setWhisperModel(const QString &path);
     QString whisperModel();
+    void setWhisperUseGpu(bool b);
+    bool whisperUseGpu() const;
 
     // Notes
     void setNotesZoom(int zoom);