Workaround for QTextToSpeech crashes if more than one instance is created/destroyed

src/calibre/gui2/tts2/config.py

@@ -125,7 +125,7 @@ class Voices(QTreeWidget):
         self.system_default_voice = Voice()
 
     def sizeHint(self) -> QSize:
-        return QSize(400, 600)
+        return QSize(400, 500)
 
     def set_voices(self, all_voices: tuple[Voice, ...], current_voice: str, engine_metadata: EngineMetadata) -> None:
         self.clear()
@@ -176,7 +176,7 @@ class EngineSpecificConfig(QWidget):
         l.addRow(_('&Output module:'), om)
         self.engine_name = ''
         om.currentIndexChanged.connect(self.rebuild_voices)
-        self.engine_instances = {}
+        self.default_output_modules = {}
         self.voice_data = {}
         self.engine_specific_settings = {}
         self.rate = r = FloatSlider(parent=self)
@@ -199,12 +199,11 @@ class EngineSpecificConfig(QWidget):
             self.engine_specific_settings[self.engine_name] = self.as_settings()
         self.engine_name = engine_name
         metadata = available_engines()[engine_name]
-        if engine_name not in self.engine_instances:
+        tts = create_tts_backend(force_engine=engine_name)
-            self.engine_instances[engine_name] = tts = create_tts_backend(force_engine=engine_name)
+        if engine_name not in self.voice_data:
             self.voice_data[engine_name] = tts.available_voices
             self.engine_specific_settings[engine_name] = EngineSpecificSettings.create_from_config(engine_name)
-        else:
+            self.default_output_modules[engine_name] = tts.default_output_module
-            tts = self.engine_instances[engine_name]
         self.output_module.blockSignals(True)
         self.output_module.clear()
         if metadata.has_multiple_output_modules:
@@ -256,7 +255,7 @@ class EngineSpecificConfig(QWidget):
         metadata = available_engines()[self.engine_name]
         output_module = self.output_module.currentData() or ''
         if metadata.has_multiple_output_modules:
-            output_module = output_module or self.engine_instances[self.engine_name].default_output_module
+            output_module = output_module or self.default_output_modules[self.engine_name].default_output_module
         all_voices = self.voice_data[self.engine_name][output_module]
         self.voices.set_voices(all_voices, s.voice_name, metadata)
 
@@ -289,7 +288,12 @@ class ConfigDialog(Dialog):
         l.addWidget(ec)
         l.addWidget(esc)
         l.addWidget(self.bb)
-        esc.set_engine(ec.value)
+        self.initial_engine_choice = ec.value
+        esc.set_engine(self.initial_engine_choice)
+
+    @property
+    def engine_changed(self) -> bool:
+        return self.engine_choice.value != self.initial_engine_choice
 
     def accept(self):
         s = self.engine_specific_config.as_settings()

src/calibre/gui2/tts2/develop.py

@@ -2,7 +2,7 @@
 # License: GPLv3 Copyright: 2024, Kovid Goyal <kovid at kovidgoyal.net>
 
 
-from qt.core import QAction, QKeySequence, QPlainTextEdit, Qt, QTextCursor, QTextToSpeech, QToolBar
+from qt.core import QAction, QKeySequence, QPlainTextEdit, QSize, Qt, QTextCursor, QTextToSpeech, QToolBar
 
 from calibre.gui2 import Application
 from calibre.gui2.main_window import MainWindow
@@ -78,7 +78,7 @@ class MainWindow(MainWindow):
                 self.display.setTextCursor(c)
         else:
             self.play_action.setChecked(True)
-        self.stop_action.setEnabled(state in (QTextToSpeech.State.Speaking, QTextToSpeech.State.Synthesizing))
+        self.stop_action.setEnabled(state in (QTextToSpeech.State.Speaking, QTextToSpeech.State.Synthesizing, QTextToSpeech.State.Paused))
 
     def toggled(self):
         if self.play_action.isChecked():
@@ -100,6 +100,9 @@ class MainWindow(MainWindow):
         c.movePosition(QTextCursor.MoveOperation.WordRight, QTextCursor.MoveMode.KeepAnchor)
         self.display.setTextCursor(c)
 
+    def sizeHint(self):
+        return QSize(500, 400)
+
 
 def main():
     app = Application([])

src/calibre/gui2/tts2/manager.py

@@ -4,7 +4,7 @@
 
 from collections import deque
 from contextlib import contextmanager
-from typing import NamedTuple, TYPE_CHECKING
+from typing import TYPE_CHECKING, NamedTuple
 
 from qt.core import QApplication, QDialog, QObject, QTextToSpeech, QWidget, pyqtSignal
 
@@ -104,6 +104,11 @@ class Tracker:
         return None
 
 
+class ResumeData:
+    is_speaking: bool = True
+    needs_full_resume: bool = False
+
+
 class TTSManager(QObject):
 
     state_changed = pyqtSignal(QTextToSpeech.State)
@@ -120,7 +125,7 @@ class TTSManager(QObject):
         if self._tts is None:
             with BusyCursor():
                 from calibre.gui2.tts2.types import create_tts_backend
-                self._tts = create_tts_backend(parent=self)
+                self._tts = create_tts_backend()
                 self._tts.state_changed.connect(self._state_changed)
                 self._tts.saying.connect(self._saying)
         return self._tts
@@ -144,12 +149,13 @@ class TTSManager(QObject):
 
     @contextmanager
     def resume_after(self):
-        is_speaking = self._tts is not None and self.state in (QTextToSpeech.State.Speaking, QTextToSpeech.State.Synthesizing, QTextToSpeech.State.Paused)
+        rd = ResumeData()
+        rd.is_speaking = self._tts is not None and self.state in (QTextToSpeech.State.Speaking, QTextToSpeech.State.Synthesizing, QTextToSpeech.State.Paused)
         if self.state is not QTextToSpeech.State.Paused:
             self.tts.pause()
-        yield is_speaking
+        yield rd
-        if is_speaking:
+        if rd.is_speaking:
-            if self._tts is None:
+            if rd.needs_full_resume:
                 self.tts.say(self.tracker.resume())
             else:
                 self.tts.resume()
@@ -162,11 +168,10 @@ class TTSManager(QObject):
         if new_rate != s.rate:
             s = s._replace(rate=new_rate)
             s.save_to_config()
-            with self.resume_after() as is_speaking:
+            with self.resume_after() as rd:
                 if self._tts is not None:
-                    if is_speaking:
+                    rd.needs_full_resume = True
-                        self.tts.stop()
+                    self.tts.reload_after_configure()
-                    self._tts = None
             return True
         return False
 
@@ -183,13 +188,16 @@ class TTSManager(QObject):
         p = self
         while p is not None and not isinstance(p, QWidget):
             p = p.parent()
-        with self.resume_after() as is_speaking:
+        with self.resume_after() as rd:
             d = ConfigDialog(parent=p)
-            if d.exec() == QDialog.DialogCode.Accepted:
+            if d.exec() == QDialog.DialogCode.Accepted and self._tts is not None:
-                if self._tts is not None:
+                rd.needs_full_resume = True
-                    if is_speaking:
+                if d.engine_changed:
+                    if rd.is_speaking:
                         self.tts.stop()
                     self._tts = None
+                else:
+                    self.tts.reload_after_configure()
 
     def _state_changed(self, state: QTextToSpeech.State) -> None:
         self.state = state

src/calibre/gui2/tts2/qt.py

@@ -11,8 +11,7 @@ class QtTTSBackend(TTSBackend):
 
     def __init__(self, engine_name: str = '', parent: QObject|None = None):
         super().__init__(parent)
-        self._voices = None
+        self._qt_reload_after_configure(engine_name)
-        self._create_engine(engine_name)
 
     @property
     def available_voices(self) -> dict[str, tuple[Voice, ...]]:
@@ -43,8 +42,14 @@ class QtTTSBackend(TTSBackend):
     def error_message(self) -> str:
         return self.tts.errorString()
 
-    def _create_engine(self, engine_name: str) -> None:
+    def reload_after_configure(self) -> None:
+        self._qt_reload_after_configure(self.engine_name)
+
+    def _qt_reload_after_configure(self, engine_name: str) -> None:
+        # Bad things happen with more than one QTextToSpeech instance
         s = {}
+        self._voices = None
+        new_backend = not hasattr(self, 'tts')
         if engine_name:
             settings = EngineSpecificSettings.create_from_config(engine_name)
             if settings.audio_device_id:
@@ -52,9 +57,15 @@ class QtTTSBackend(TTSBackend):
                     if bytes(x.id) == settings.audio_device_id.id:
                         s['audioDevice'] = x
                         break
+            if new_backend:
                 self.tts = QTextToSpeech(engine_name, s, self)
             else:
+                self.tts.setEngine(engine_name, s)
+        else:
+            if new_backend:
                 self.tts = QTextToSpeech(self)
+            else:
+                self.tts.setEngine('')
             engine_name = self.tts.engine()
             settings = EngineSpecificSettings.create_from_config(engine_name)
             if settings.audio_device_id:
@@ -63,6 +74,9 @@ class QtTTSBackend(TTSBackend):
                         s['audioDevice'] = x
                         self.tts = QTextToSpeech(engine_name, s, self)
                         break
+        if new_backend:
+            self.tts.sayingWord.connect(self._saying_word)
+            self.tts.stateChanged.connect(self._state_changed)
 
         self.tts.setRate(max(-1, min(float(settings.rate), 1)))
         self.tts.setPitch(max(-1, min(float(settings.pitch), 1)))
@@ -73,9 +87,10 @@ class QtTTSBackend(TTSBackend):
                 if v.name() == settings.voice_name:
                     self.tts.setVoice(v)
                     break
-        self.tts.sayingWord.connect(self._saying_word)
-        self.tts.stateChanged.connect(self.state_changed.emit)
         self._current_settings = settings
 
     def _saying_word(self, word: str, utterance_id: int, start: int, length: int) -> None:
         self.saying.emit(start, length)
+
+    def _state_changed(self, state: QTextToSpeech.State) -> None:
+        self.state_changed.emit(state)

src/calibre/gui2/tts2/types.py

@@ -7,7 +7,7 @@ from enum import Enum, auto
 from functools import lru_cache
 from typing import Literal, NamedTuple
 
-from qt.core import QLocale, QObject, QTextToSpeech, QVoice, pyqtSignal
+from qt.core import QApplication, QLocale, QObject, QTextToSpeech, QVoice, pyqtSignal
 
 from calibre.constants import islinux, ismacos, iswindows
 from calibre.utils.config import JSONConfig
@@ -208,20 +208,30 @@ class TTSBackend(QObject):
     def error_message(self) -> str:
         raise NotImplementedError()
 
+    def reload_after_configure(self) -> str:
+        raise NotImplementedError()
 
-def create_tts_backend(parent: QObject|None = None, force_engine: str | None = None) -> TTSBackend:
+
+engine_instances: dict[str, TTSBackend] = {}
+
+
+def create_tts_backend(force_engine: str | None = None) -> TTSBackend:
     prefs = load_config()
     engine_name = prefs.get('engine', '') if force_engine is None else force_engine
     engine_name = engine_name or default_engine_name()
     if engine_name not in available_engines():
         engine_name = default_engine_name()
-
     if engine_name == 'speechd':
+        if engine_name not in engine_instances:
             from calibre.gui2.tts2.speechd import SpeechdTTSBackend
-        ans = SpeechdTTSBackend(engine_name, parent)
+            engine_instances[engine_name] = SpeechdTTSBackend(engine_name, QApplication.instance())
+        ans = engine_instances[engine_name]
     else:
-        if engine_name not in available_engines():
+        if 'qt' not in engine_instances:
-            engine_name = ''  # let Qt pick the engine
+            # Bad things happen with more than one QTextToSpeech instance
             from calibre.gui2.tts2.qt import QtTTSBackend
-        ans = QtTTSBackend(engine_name, parent)
+            engine_instances['qt'] = QtTTSBackend(engine_name if engine_name in available_engines() else '', QApplication.instance())
+        ans = engine_instances['qt']
+        if ans.engine_name != engine_name:
+            ans._qt_reload_after_configure(engine_name if engine_name in available_engines() else '')
     return ans