WIP: parakeet

stt_voice_messages/README.md

@@ -1,28 +1,49 @@
-# Requirements
+# About
 
-## STT Models
+This plugin allows you in conjuction with a _general-purpose speech recognition model_ to transcribe your voice messages to text.
 
-### openai-whisper https://github.com/openai/whisper
+In order to make use of this plugin, you need to have at least one of the following models installed:
 
-#### Installation
-`pip install -U openai-whisper` will install
+#### OpenAI Whisper
+- Website: https://github.com/openai/whisper
+- Installable by: `pip install -U openai-whisper`
 
-```
-mpmath, urllib3, tqdm, sympy, regex, nvidia-nvtx-cu12, nvidia-nvjitlink-cu12,
-nvidia-nccl-cu12, nvidia-curand-cu12, nvidia-cufft-cu12, nvidia-cuda-runtime-cu12,
-nvidia-cuda-nvrtc-cu12, nvidia-cuda-cupti-cu12, nvidia-cublas-cu12, networkx,
-MarkupSafe, llvmlite, fsspec, filelock, charset-normalizer, certifi, triton,
-requests, nvidia-cusparse-cu12, nvidia-cudnn-cu12, numba, jinja2, tiktoken,
-nvidia-cusolver-cu12, torch, openai-whisper
-```
+#### Faster Whisper
+- Website: https://github.com/SYSTRAN/faster-whisper
+- Installable by: `pip install -U faster-whisper`
 
-#### Models
+Additionally you have to checkout the following Gajim branch:
+https://dev.gajim.org/mesonium/gajim/-/tree/stt_voice_messages
 
-| Multi Langual Model | Download Size | VRAM Requirement | Relative Speed |
-|---------------------|---------------| ---------------- |----------------|
-| Tiny                | 70 MB         | ~1 GB | ~32x           |
-| Base                | 140 MB        | ~1 GB | ~16x           | 
-| Small               | 460 MB        | ~2 GB | ~6x            |
-| Medium              | 1.4 GB        | ~5 GB | ~2x            |
-| Large               | 2.9 GB        | ~10 GB | ~1x            |
+# Hint
 
+_**The plugin is very much POC at this stage!**_
+
+Currently a chosen model will be on first downloaded in the background, during which 
+Gajim's UI may not respond.
+
+Typical model sizes are in case of OpenAI Whisper:
+
+| Multi Langual Model | Download Size | 
+|---------------------|---------------| 
+| Tiny                | 70 MB         |
+| Base                | 140 MB        | 
+| Small               | 460 MB        |
+| Medium              | 1.4 GB        | 
+| Large               | 2.9 GB        | 
+
+# TODO
+
+- [x] Offer multiple models
+- [ ] Add various model settings
+- [ ] Model receiving
+  - [ ] Hint model download state
+  - [ ] Allow to change model download location
+  - [ ] Allow to use local models
+- [ ] Database Handling
+  - [ ] Store transcribed messages in a DB
+  - [ ] Option to delete DB
+- [ ] Update UI
+  - [ ] Make it more pretty
+  - [ ] Show progress bar
+  - [ ] Highlight words on playback

stt_voice_messages/gtk/config_dialog.py

@@ -18,22 +18,23 @@ from __future__ import annotations
 import logging
 import typing
 from pathlib import Path
-from typing import TYPE_CHECKING
+from typing import TYPE_CHECKING, Any
 
-import whisper
-from gi.repository import Gtk
+try:
+    import onnx_asr
+except ModuleNotFoundError:
+    if typing.TYPE_CHECKING:
+        import onnx_asr
+
+from gi.repository import Adw, Gtk
 
-from gajim.common import app
-from gajim.common.app import Any
-from gajim.gtk.builder import get_builder
 from gajim.gtk.const import Setting, SettingKind, SettingType
-from gajim.gtk.settings import SettingsBox
-from gajim.gtk.sidebar_switcher import SideBarSwitcher
-from gajim.plugins.helpers import get_builder
+from gajim.gtk.filechoosers import Filter
+from gajim.gtk.settings import GajimPreferencesGroup, SettingsDialog
 from gajim.plugins.plugins_i18n import _
 
-from ..model_settings import *
-from ..models import openai_whisper
+from ..models import stt
+from ..models.model_settings import OnnxAsrSettings
 
 if TYPE_CHECKING:
     from ..stt_voice_messages import STTVoiceMessagesPlugin
@@ -41,210 +42,250 @@ if TYPE_CHECKING:
 log = logging.getLogger('gajim.p.sttvm_config_dialog')
 
 
-SUPPORTED_MODELS: dict[str, dict[str, typing.Union[list[str], Any, str]]] = {
-    'model_openaiwhisper': {
-        'moduls': ['whisper'],
-        'class': openai_whisper.WhisperModel,
-        'name': 'OpenAI Whisper'
-    },
-    'model_ctranslate2': {
-        'moduls': ['ctranslate2'],
-        'class': None,
-        'name': _('CTranslate2')
-    },
-    'model_faster-whisper': {
-        'moduls': ['faster-whisper'],
-        'class': None,
-        'name:': _('Faster-Whisper')
-    },
-    'model_distill': {
-        'moduls': ['transformers', 'accelerate', 'datasets[audio]'],
-        'class': None,
-        'name': _('Distill')
-    }
-}
-
-
 class Configuration:
     def __init__(self, plugin: STTVoiceMessagesPlugin):
         self._plugin = plugin
-        self._openaiwhisper_settings = OpenAIWhisperSettings()
-        self._available_models: dict[
-            str, dict[str, typing.Union[list[str], Any, str]]] = {}
-        self.check_available_moduls()
-            
+        self._instance = None
+        self._main_model_row = None
+        self._preset_model_picker = None
+        self._custom_model_id_entry = None
+        self._local_model_file_picker = None
+        self._status_group = None
+        self._model_data: dict[str, str] = {}
+        self._instance = stt.OnnxAsrModel()
+        self._instance.set_config(OnnxAsrSettings(
+            model_id=self.plugin.config['model_id'],
+            model_path=self.plugin.config['model_path']
+        ))
+        self._model_data = self._steal_model_list()
+
     @property
     def plugin(self) -> STTVoiceMessagesPlugin:
         return self._plugin
 
     @property
-    def available_models(self) -> dict[str, dict[str, typing.Union[list[str], Any, str]]]:
-        return self._available_models
+    def is_available(self) -> bool:
+        return self._instance is not None
+
+    def unload_model(self) -> None:
+        if self._instance is not None:
+            self._instance.unload_now()
+
+    def _steal_model_list(self) -> dict[str, str]:
+        # UGLY: Extract available model choices from onnx_asr type hints.
+        ann = onnx_asr.load_model.__annotations__.get('model')
+        return {
+            v: v for arg in typing.get_args(ann)
+            for v in typing.get_args(arg)
+            if isinstance(v, str)
+        }
 
     def on_setting(self, value: Any, data: Any) -> None:
         if isinstance(value, str):
-            value.strip()
-        log.debug('plugin config before:\n %s', self.plugin.config.data)
+            value = value.strip()
         self.plugin.config[data] = value
-        self._plugin.config['model_instance'].on_setting(data, value)
-        log.debug('plugin config after:\n %s', self.plugin.config.data)
 
-    def on_set_model(self, value: Any, data: Any) -> None:
-        if isinstance(value, str):
-            value.strip()
-        log.debug('plugin config before:\n %s', self.plugin.config.data)
+    def on_preset_changed(self, value: str, data: Any) -> None:
+        if self._custom_model_id_entry is not None:
+            entry_text = self._custom_model_id_entry.entry.get_text().strip()
+            if entry_text:
+                self._update_model_status()
+                return  # custom entry overrides; ignore preset change
+        self._write_model_id(value)
+        self._update_model_status()
 
-        self._available_models[value]['model_instance'] = self._available_models[value]['class']()
+    def on_custom_model_id_changed(self, value: str, data: Any) -> None:
+        value = value.strip()
+        if value:
+            self._write_model_id(value)
+        elif self._preset_model_picker is not None:
+            preset_key = self._preset_model_picker._dropdown.get_selected_key()
+            if preset_key is not None:
+                self._write_model_id(preset_key)
+        self._apply_sensitivity_state()
+        self._update_model_status()
 
-        self.plugin.config['model_class'] = self._available_models[value][
-            'class']
-        self.plugin.config['model_instance'] = self._available_models[value]['model_instance']
+    def on_model_file_picked(self, value: str, data: Any) -> None:
+        self._write_model_path(str(Path(value).parent) if value else '')
+        self._apply_sensitivity_state()
+        self._update_model_status()
 
-        self.on_setting(value, data)
-        log.debug('plugin config after:\n %s', self.plugin.config.data)
+    def _write_model_id(self, model_id: str) -> None:
+        if self.plugin.config['model_id'] == model_id:
+            return
+        self.plugin.config['model_id'] = model_id
+        if self._instance is not None:
+            self._instance.set_config(OnnxAsrSettings(
+                model_id=self.plugin.config['model_id'],
+                model_path=self.plugin.config['model_path']
+            ))
 
-    @staticmethod
-    def is_module_available(module: str) -> bool:
-        try:
-            __import__(module)
-            return True
-        except ModuleNotFoundError:
-            log.debug('Could not find module %s', module)
-            return False
-        except ImportError as ex:
-            log.debug(str(ex))
-            return False
+    def _write_model_path(self, model_path: str) -> None:
+        if self.plugin.config['model_path'] == model_path:
+            return
+        self.plugin.config['model_path'] = model_path
+        if self._instance is not None:
+            self._instance.set_config(OnnxAsrSettings(
+                model_id=self.plugin.config['model_id'],
+                model_path=self.plugin.config['model_path']
+            ))
 
-    def check_available_moduls(self):
-        for model in SUPPORTED_MODELS:
-            available = True
-            for modul in SUPPORTED_MODELS[model]['moduls']:
-                if not self.is_module_available(modul):
-                    available = False
-                    continue
-            if available:
-                self._available_models[model] = SUPPORTED_MODELS[model]
+    def sync_model_path_from_widget(self) -> None:
+        if self._local_model_file_picker is None:
+            return
+        button = self._local_model_file_picker.get_activatable_widget()
+        path = button.get_path()
+        new_path = str(path.parent) if path else ''
+        self._write_model_path(new_path)
 
-        if (self.plugin.config.data['model_class'] is None
-                and len(self._available_models) > 0):
-            model = list(self._available_models)[0]
-            self.on_set_model(model, 'model')
-            log.debug('Choose first available model!')
+    def _apply_sensitivity_state(self) -> None:
+        if self._preset_model_picker is None:
+            return
+        has_local = bool(self.plugin.config['model_path'])
+        entry_text = (self._custom_model_id_entry.entry.get_text().strip()
+                      if self._custom_model_id_entry else '')
+        has_entry = bool(entry_text)
+        self._custom_model_id_entry.set_sensitive(not has_local)
+        self._preset_model_picker.set_sensitive(not has_local and not has_entry)
+
+    def _update_model_status(self) -> None:
+        if self._main_model_row is None:
+            return
+        entry_text = (self._custom_model_id_entry.entry.get_text().strip()
+                      if self._custom_model_id_entry else '')
+
+        if self.plugin.config['model_path']:
+            path = Path(self.plugin.config['model_path'])
+            summary = _('Local: {}').format(path.name or str(path))
+            description = _('Loading model files from {}').format(path)
+            if not (path / 'config.json').exists():
+                description += '\n' + _(
+                    'config.json not found in this directory — onnx-asr will'
+                    ' fall back to Model preset or Custom Model ID for the'
+                    ' architecture.')
+        elif entry_text:
+            summary = _('Custom: {}').format(entry_text)
+            description = _('Using custom model: {}').format(entry_text)
         else:
-            log.debug('Available model already chosen!')
+            preset_key = (self._preset_model_picker._dropdown.get_selected_key()
+                          if self._preset_model_picker else '')
+            summary = preset_key or _('(none)')
+            description = (_('Using preset: {}').format(preset_key)
+                           if preset_key else '')
 
-        log.debug('models = %s', self._available_models)
+        self._main_model_row._label.set_text(summary)
+        if self._status_group is not None:
+            self._status_group.set_description(description)
 
 
-
-class PreferenceBox(SettingsBox):
-    def __init__(self, settings: list[Setting]) -> None:
-        SettingsBox.__init__(self, None)
-        self.get_style_context().add_class('border')
-        self.set_selection_mode(Gtk.SelectionMode.NONE)
-        self.set_vexpand(False)
-        self.set_valign(Gtk.Align.END)
-
-        for setting in settings:
-            self.add_setting(setting)
-        self.update_states()
-
-
-class STTVoiceMessagesConfigDialog(Gtk.ApplicationWindow):
+class STTVoiceMessagesConfigDialog(SettingsDialog):
     def __init__(self, config: Configuration, parent: Gtk.Window) -> None:
-        Gtk.ApplicationWindow.__init__(self)
-
-        self.set_application(app.app)
-        self.set_position(Gtk.WindowPosition.CENTER)
-        self.set_show_menubar(False)
-        self.set_name('PreferencesWindow')
-        self.set_default_size(900, 650)
-        self.set_resizable(True)
-        self.set_title(_('STT Voice Messages - Preferences'))
-
-        ui_path = Path(__file__).parent
-        self._ui = get_builder(str(ui_path.resolve() / 'config_dialog.ui'))
-
-        self._prefs: dict[str, PreferenceBox] = {}
-        side_bar_switcher = SideBarSwitcher()
-        side_bar_switcher.set_stack(self._ui.stack)
-        self._ui.grid.attach(side_bar_switcher, 0, 0, 1, 1)
-        self.add(self._ui.grid)
-
         self.config = config
         self.plugin = self.config.plugin
+        if not config.is_available:
+            return
 
-        prefs: list[tuple[str, type[PreferenceBox]]] = [
-            ('stt_behaviour', self.STTBehaviour),
-            ('models', self.Models),
-            ('whisper_general', self.OpenAIWhisperGeneral),
+        rows = [
+            Setting(SettingKind.SWITCH,
+                    _('Auto Transcribe'),
+                    SettingType.VALUE,
+                    value=self.plugin.config['auto_transcribe'],
+                    data='auto_transcribe',
+                    callback=config.on_setting,
+                    desc=_('Transcribe messages as they appear')),
+            Setting(SettingKind.SUBPAGE,
+                    _('Model'),
+                    SettingType.VALUE,
+                    value=None,
+                    name='main_model',
+                    props={'subpage': 'sttvm-model'}),
         ]
-        self._add_prefs(prefs)
 
-        self.show_all()
+        SettingsDialog.__init__(
+            self,
+            parent,
+            _('STT Voice Messages'),
+            Gtk.DialogFlags.MODAL,
+            rows,
+            '',
+        )
 
-    class STTBehaviour(PreferenceBox):
-        def __init__(self, config_dialog: STTVoiceMessagesConfigDialog) -> None:
+        config._main_model_row = self.get_setting('main_model')
 
-            settings = [
-                Setting(SettingKind.SWITCH,
-                        _('Auto Transcribe'),
-                        SettingType.VALUE,
-                        value=config_dialog.plugin.config['auto_transcribe'],
-                        data='auto_transcribe',
-                        callback=config_dialog.config.on_setting)
-            ]
-
-            PreferenceBox.__init__(self, settings)
-
-    class Models(PreferenceBox):
-        def __init__(self, config_dialog: STTVoiceMessagesConfigDialog) -> None:
-            models: list[tuple[str, str]] = []
-            for key, value in config_dialog.config.available_models.items():
-                models.append(
-                    (key, str(value['name']))
-                )
-
-            settings = [
-                Setting(SettingKind.COMBO,
-                        _('Speech To Text Model'),
-                        SettingType.VALUE,
-                        value=config_dialog.plugin.config['model'],
-                        data='model',
-                        callback=config_dialog.config.on_set_model,
-                        props={'combo_items': models},
-                        desc=_('Choose Model to use')),
-            ]
-
-            PreferenceBox.__init__(self, settings)
-
-    class OpenAIWhisperGeneral(PreferenceBox):
-        def __init__(self, config_dialog: STTVoiceMessagesConfigDialog) -> None:
-
-            settings = [
-                Setting(SettingKind.POPOVER,
-                        _('Language Model Size'),
-                        SettingType.VALUE,
-                        value=config_dialog.plugin.config['whisperai_model_size'],
-                        data='whisperai_model_size',
-                        callback=config_dialog.config.on_setting,
-                        props={'entries': whisper.available_models()}),
-
-                Setting(SettingKind.SWITCH,
-                        _('Translate'),
-                        SettingType.VALUE,
-                        value=config_dialog.plugin.config['whisperai_translate'],
-                        data='whisperai_translate',
-                        callback=config_dialog.config.on_setting)
-            ]
-
-            PreferenceBox.__init__(self, settings)
-
-    def _add_prefs(self, prefs: list[tuple[str, type[PreferenceBox]]]):
-        for ui_name, klass in prefs:
-            pref_box = getattr(self._ui, ui_name)
-            pref = klass(self)  # pyright: ignore
-            pref_box.add(pref)
-            self._prefs[ui_name] = pref
+        use_custom = self.plugin.config['model_id'] not in config._model_data
 
 
+        subpage_rows: list[Setting] = [
+            Setting(SettingKind.DROPDOWN,
+                    _('Model'),
+                    SettingType.VALUE,
+                    value=self.plugin.config['model_id'],
+                    name='preset_model',
+                    callback=config.on_preset_changed,
+                    props={'data': config._model_data}),
+            Setting(SettingKind.ENTRY,
+                    _('Custom Model'),
+                    SettingType.VALUE,
+                    value=self.plugin.config['model_id'] if use_custom else '',
+                    name='custom_model',
+                    callback=config.on_custom_model_id_changed,
+                    desc=_('Custom HF model path or model ID')),
+            Setting(SettingKind.FILECHOOSER,
+                    _('Local File'),
+                    SettingType.VALUE,
+                    value='',
+                    name='local_model_file',
+                    callback=config.on_model_file_picked,
+                    desc=_('Model ID is taken from config.json if not set'),
+                    props={'filefilters': [
+                        Filter(_('ONNX model'), suffixes=['onnx'], default=True),
+                    ]}),
+        ]
+
+        controls_group = GajimPreferencesGroup('model_controls')
+        for s in subpage_rows:
+            controls_group.add_setting(s)
+
+        status_group = Adw.PreferencesGroup()
+
+        pref_page = Adw.PreferencesPage()
+        pref_page.add(controls_group)
+        pref_page.add(status_group)
+
+        toolbar = Adw.ToolbarView(content=pref_page)
+        toolbar.add_top_bar(Adw.HeaderBar())
+
+        page = Adw.NavigationPage(
+            tag='sttvm-model', title=_('Model'), child=toolbar)
+        self._nav.add(page)
+
+        config._preset_model_picker = controls_group.get_setting('preset_model')
+        config._custom_model_id_entry = controls_group.get_setting('custom_model')
+        config._local_model_file_picker = controls_group.get_setting(
+            'local_model_file')
+        config._status_group = status_group
+
+        config._custom_model_id_entry.entry.set_placeholder_text(
+            _('onnx-community/whisper-large-v3-turbo'))
+
+        button = config._local_model_file_picker.get_activatable_widget()
+        button._label_text = _('.oonx')
+        button.reset()
+
+        if self.plugin.config['model_path']:
+            onnx_in_dir = next(iter(Path(self.plugin.config['model_path']).glob('*.onnx')),
+                               None)
+            if onnx_in_dir is not None:
+                button.set_path(onnx_in_dir)
+
+        config._update_model_status()
+        config._apply_sensitivity_state()
+
+    def _cleanup(self) -> None:
+        self.config.sync_model_path_from_widget()
+        self.config._main_model_row = None
+        self.config._preset_model_picker = None
+        self.config._custom_model_id_entry = None
+        self.config._local_model_file_picker = None
+        self.config._status_group = None
+        SettingsDialog._cleanup(self)

stt_voice_messages/gtk/config_dialog.ui

@@ -1,271 +0,0 @@
-<?xml version="1.0" encoding="UTF-8"?>
-<!-- Generated with glade 3.40.0 -->
-<interface>
-  <requires lib="gtk+" version="3.20"/>
-  <!-- n-columns=3 n-rows=3 -->
-  <object class="GtkGrid" id="grid">
-    <property name="visible">True</property>
-    <property name="can-focus">False</property>
-    <child>
-      <object class="GtkStack" id="stack">
-        <property name="visible">True</property>
-        <property name="can-focus">False</property>
-        <property name="hexpand">True</property>
-        <child>
-          <object class="GtkScrolledWindow">
-            <property name="visible">True</property>
-            <property name="can-focus">True</property>
-            <property name="hscrollbar-policy">never</property>
-            <property name="shadow-type">in</property>
-            <property name="overlay-scrolling">False</property>
-            <child>
-              <object class="GtkViewport">
-                <property name="visible">True</property>
-                <property name="can-focus">False</property>
-                <child>
-                  <object class="GtkBox">
-                    <property name="visible">True</property>
-                    <property name="can-focus">False</property>
-                    <property name="orientation">vertical</property>
-                    <property name="spacing">24</property>
-                    <child>
-                      <!-- n-columns=1 n-rows=1 -->
-                      <object class="GtkGrid" id="stt_behaviour">
-                        <property name="visible">True</property>
-                        <property name="can-focus">False</property>
-                        <property name="orientation">vertical</property>
-                        <property name="row-spacing">12</property>
-                        <child>
-                          <object class="GtkLabel">
-                            <property name="visible">True</property>
-                            <property name="can-focus">False</property>
-                            <property name="label" translatable="yes">Behaviour of STT Voice Messages</property>
-                            <property name="xalign">0</property>
-                            <style>
-                              <class name="bold"/>
-                            </style>
-                          </object>
-                          <packing>
-                            <property name="left-attach">0</property>
-                            <property name="top-attach">0</property>
-                          </packing>
-                        </child>
-                      </object>
-                      <packing>
-                        <property name="expand">False</property>
-                        <property name="fill">True</property>
-                        <property name="position">0</property>
-                      </packing>
-                    </child>
-                    <child>
-                      <!-- n-columns=1 n-rows=1 -->
-                      <object class="GtkGrid" id="models">
-                        <property name="visible">True</property>
-                        <property name="can-focus">False</property>
-                        <property name="orientation">vertical</property>
-                        <property name="row-spacing">12</property>
-                        <child>
-                          <object class="GtkLabel">
-                            <property name="visible">True</property>
-                            <property name="can-focus">False</property>
-                            <property name="label" translatable="yes">General Model Configuration</property>
-                            <property name="xalign">0</property>
-                            <style>
-                              <class name="bold"/>
-                            </style>
-                          </object>
-                          <packing>
-                            <property name="left-attach">0</property>
-                            <property name="top-attach">0</property>
-                          </packing>
-                        </child>
-                      </object>
-                      <packing>
-                        <property name="expand">False</property>
-                        <property name="fill">True</property>
-                        <property name="position">1</property>
-                      </packing>
-                    </child>
-                    <child>
-                      <!-- n-columns=1 n-rows=1 -->
-                      <object class="GtkGrid" id="file_preview">
-                        <property name="visible">True</property>
-                        <property name="can-focus">False</property>
-                        <property name="orientation">vertical</property>
-                        <property name="row-spacing">12</property>
-                        <child>
-                          <object class="GtkLabel">
-                            <property name="visible">True</property>
-                            <property name="can-focus">False</property>
-                            <property name="label" translatable="yes">Preview UI</property>
-                            <property name="xalign">0</property>
-                            <style>
-                              <class name="bold"/>
-                            </style>
-                          </object>
-                          <packing>
-                            <property name="left-attach">0</property>
-                            <property name="top-attach">0</property>
-                          </packing>
-                        </child>
-                      </object>
-                      <packing>
-                        <property name="expand">False</property>
-                        <property name="fill">True</property>
-                        <property name="position">2</property>
-                      </packing>
-                    </child>
-                  </object>
-                </child>
-              </object>
-            </child>
-          </object>
-          <packing>
-            <property name="name">general</property>
-            <property name="title" translatable="yes">General</property>
-            <property name="icon-name">computer-symbolic</property>
-          </packing>
-        </child>
-        <child>
-          <object class="GtkScrolledWindow">
-            <property name="visible">True</property>
-            <property name="can-focus">True</property>
-            <property name="hscrollbar-policy">never</property>
-            <property name="shadow-type">in</property>
-            <property name="overlay-scrolling">False</property>
-            <child>
-              <object class="GtkViewport">
-                <property name="visible">True</property>
-                <property name="can-focus">False</property>
-                <child>
-                  <object class="GtkBox">
-                    <property name="visible">True</property>
-                    <property name="can-focus">False</property>
-                    <property name="orientation">vertical</property>
-                    <property name="spacing">24</property>
-                    <child>
-                      <!-- n-columns=3 n-rows=3 -->
-                      <object class="GtkGrid" id="whisper_general">
-                        <property name="visible">True</property>
-                        <property name="can-focus">False</property>
-                        <property name="orientation">vertical</property>
-                        <property name="row-spacing">12</property>
-                        <child>
-                          <object class="GtkLabel">
-                            <property name="visible">True</property>
-                            <property name="can-focus">False</property>
-                            <property name="label" translatable="yes">General</property>
-                            <property name="xalign">0</property>
-                            <style>
-                              <class name="bold"/>
-                            </style>
-                          </object>
-                          <packing>
-                            <property name="left-attach">0</property>
-                            <property name="top-attach">0</property>
-                          </packing>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                        <child>
-                          <placeholder/>
-                        </child>
-                      </object>
-                      <packing>
-                        <property name="expand">False</property>
-                        <property name="fill">True</property>
-                        <property name="position">0</property>
-                      </packing>
-                    </child>
-                    <child>
-                      <placeholder/>
-                    </child>
-                    <child>
-                      <placeholder/>
-                    </child>
-                    <child>
-                      <placeholder/>
-                    </child>
-                    <child>
-                      <placeholder/>
-                    </child>
-                  </object>
-                </child>
-              </object>
-            </child>
-          </object>
-          <packing>
-            <property name="name">openai-whisper</property>
-            <property name="title" translatable="yes">openAI Whisper</property>
-            <property name="position">1</property>
-          </packing>
-        </child>
-        <child>
-          <placeholder/>
-        </child>
-        <child>
-          <placeholder/>
-        </child>
-        <child>
-          <placeholder/>
-        </child>
-        <child>
-          <placeholder/>
-        </child>
-        <child>
-          <placeholder/>
-        </child>
-        <style>
-          <class name="settings-stack"/>
-        </style>
-      </object>
-      <packing>
-        <property name="left-attach">1</property>
-        <property name="top-attach">0</property>
-      </packing>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-    <child>
-      <placeholder/>
-    </child>
-  </object>
-</interface>

stt_voice_messages/gtk/sttbox.py

@@ -13,9 +13,12 @@
 # You should have received a copy of the GNU General Public License
 # along with Gajim. If not, see <http://www.gnu.org/licenses/>.
 
-import logging
+from __future__ import annotations
 
-from gi.repository import Gtk
+import logging
+from pathlib import Path
+
+from gi.repository import Gtk, Adw
 
 from gajim.plugins.gajimplugin import GajimPluginConfig
 from gajim.plugins.plugins_i18n import _
@@ -26,48 +29,62 @@ log = logging.getLogger('gajim.p.stt_voice_messages_sttbox')
 
 class STTBox(Gtk.Box):
     def __init__(self,
-                 preview_audio_widget: Gtk.Box,
                  config: GajimPluginConfig,
-                 audio_file: str,
+                 audio_file: Path,
                  ) -> None:
 
-        Gtk.Box.__init__(self, orientation=Gtk.Orientation.VERTICAL, spacing=12)
+        Gtk.Box.__init__(self, orientation=Gtk.Orientation.HORIZONTAL, spacing=6)
 
         self._config = config
-        self._preview_audio = preview_audio_widget
         self._model = None
         self._audio_file = audio_file
         self._text = ''
 
-        self._transcribe_button = Gtk.Button(label=_('Transcribe'))
+        self._transcribe_button = Gtk.Button.new_from_icon_name("lucide-captions-symbolic")
+        self._transcribe_button.set_tooltip_text(_('Transcribe voice message'))
+
+        self._spinner = Adw.Spinner(valign=Gtk.Align.START, visible=False)
 
         self._transcription_label = Gtk.Label(
             label=_('Nothing transcribed yet'))
         self._transcription_label.set_max_width_chars(40)
-        self._transcription_label.set_line_wrap(True)
-
-        self.add(self._transcribe_button)
-        self.add(self._transcription_label)
+        self._transcription_label.set_wrap(True)
 
+        self.append(self._spinner)
+        self.append(self._transcription_label)
+        self._transcribe_button.connect('clicked', self._on_transcribe_clicked)
         self._result = helper.Results('')
 
-        self._transcribe_button.connect('clicked', self._on_transcribe_clicked)
-
-        self.show_all()
+    @property
+    def button(self) -> Gtk.Button:
+        return self._transcribe_button
 
     def _on_transcribe_clicked(self, _button: Gtk.Button) -> None:
-        log.debug('config.data = %s', self._config.data)
-        model = self._config.data['model_instance']
-        if model is None:
+        log.debug('config._instance = %s', self._config._instance)
+        self._model = self._config._instance
+        if self._model is None:
             return
 
-        self._model = model
+        if self._model.is_loaded:
+            text = _('Transcribing…')
+        elif self._model.will_download:
+            text = _('Downloading ') + self._model.model_id
+        else:
+            text = _('Loading model…')
+        self._transcription_label.set_text(text)
+        self._spinner.set_visible(True)
+        self._task = helper.BackgroundTask(
+            self._model.load, self._on_load_done)
+        self._task.start()
 
-        transcription_task = helper.BackgroundTask(
-            self._model.transcribe(self._result, self._audio_file),
-            self._show_result
+    def _on_load_done(self):
+        self._transcription_label.set_text(_('Transcribing…'))
+        self._task = helper.BackgroundTask(
+            lambda: self._model.recognize(
+                self._result, helper.load_audio(self._audio_file)),
+            self._show_result,
         )
-        transcription_task.start()
+        self._task.start()
 
     def _show_result(self):
         assert self._model is not  None
@@ -76,3 +93,4 @@ class STTBox(Gtk.Box):
             self._transcription_label.set_text(self._text.strip())
         else:
             self._transcription_label.set_text(_('_Have not heard any word!_'))
+        self._spinner.set_visible(False)

stt_voice_messages/helper.py

@@ -13,16 +13,53 @@
 # You should have received a copy of the GNU General Public License
 # along with Gajim. If not, see <http://www.gnu.org/licenses/>.
 
+import logging
+import typing
 from dataclasses import dataclass
+from pathlib import Path
 
+import gi
+import numpy as np
 from gi.repository import Gio, GObject
 
+try:
+    gi.require_version('Gst', '1.0')
+    from gi.repository import Gst
+except Exception:
+    if typing.TYPE_CHECKING:
+        from gi.repository import Gst
+
+log = logging.getLogger('gajim.p.sttvm_helper')
+
 
 @dataclass
 class Results:
     text: str
 
 
+def load_audio(path: Path, sample_rate: int = 16000) -> np.ndarray:
+    Gst.init(None)
+    pipeline = Gst.parse_launch(
+        'filesrc name=src ! decodebin ! audioconvert ! audioresample ! '
+        f'audio/x-raw,format=F32LE,rate={sample_rate},channels=1 ! '
+        'appsink name=sink sync=false'
+    )
+    pipeline.get_by_name('src').set_property('location', str(path))
+    sink = pipeline.get_by_name('sink')
+    chunks: list[np.ndarray] = []
+
+    pipeline.set_state(Gst.State.PLAYING)
+    while (sample := sink.emit('try-pull-sample', 10 * Gst.SECOND)) is not None:
+        buf = sample.get_buffer()
+        _, info = buf.map(Gst.MapFlags.READ)
+        chunks.append(np.frombuffer(bytes(info.data), dtype=np.float32))
+        buf.unmap(info)
+    pipeline.set_state(Gst.State.NULL)
+
+    if not chunks:
+        raise RuntimeError(f'Could not decode audio: {path}')
+    return np.concatenate(chunks)
+
 
 '''
 https://discourse.gnome.org/t/gtk-threading-problem-with-glib-idle-add/13597/5
@@ -57,6 +94,7 @@ class BackgroundTask(GObject.Object):
             retval = self.function()
             task.return_value(retval)
         except Exception as e:
+            log.exception('Background task failed')
             task.return_value(e)
 
     def finish(self):

stt_voice_messages/models/distill.py

@@ -18,7 +18,7 @@ from pathlib import Path
 
 from gajim.gtk.const import Setting
 
-from .model import Model
+from .model_template import Model
 
 try:
     import ctranslate2

stt_voice_messages/models/faster_whisper.py

@@ -0,0 +1,60 @@
+# This file is part of Gajim.
+#
+# Gajim is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Gajim is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Gajim. If not, see <http://www.gnu.org/licenses/>.
+
+import logging
+import typing
+from dataclasses import dataclass
+from pathlib import Path
+
+from ..helper import Results
+from .model_settings import FasterWhisperSettings
+from .model_template import Model
+
+log = logging.getLogger('gajim.p.sttvm_faster_whisper')
+
+try:
+    import faster_whisper
+except ModuleNotFoundError:
+    if typing.TYPE_CHECKING:
+        import faster_whisper
+
+@dataclass
+class Configuration:
+    model_size: str
+
+class FasterWhisperModel(Model):
+    def __init__(self):
+        self._result: str = ''
+        self._config = FasterWhisperSettings()
+
+    @property
+    def result(self) -> str:
+        return self._result
+
+    def transcribe(self, result: Results, audio_file: Path) -> None:
+        model = faster_whisper.WhisperModel(self._config.model_size, compute_type="float32")
+        log.debug('model size is used = %s', self._config.model_size)
+        segments, _ = model.transcribe(audio_file)
+        segments = list(segments)
+        result.text = ''
+        log.debug('segments = %s', segments)
+        for segment in segments:
+            result.text += segment.text
+            print("[%.2fs -> %.2fs] %s" % (
+            segment.start, segment.end, segment.text))
+
+    def set_config(self, config: FasterWhisperSettings) -> None:
+        self._config = config
+

stt_voice_messages/models/model_settings.py

@@ -18,6 +18,6 @@ from dataclasses import dataclass, field
 
 
 @dataclass
-class OpenAIWhisperSettings:
-    whisperai_model_size: str = field(default='tiny', init=True)
-
+class OnnxAsrSettings:
+    model_id: str = field(default='nemo-parakeet-tdt-0.6b-v3', init=True)
+    model_path: str = ''

stt_voice_messages/models/model_template.py

@@ -14,19 +14,28 @@
 # along with Gajim. If not, see <http://www.gnu.org/licenses/>.
 
 from abc import ABC, abstractmethod
-from pathlib import Path
+from typing import Any
 
-from gajim.gtk.const import Setting
+import numpy as np
 
 from ..helper import Results
 
 
 class Model(ABC):
 
+    @property
     @abstractmethod
-    def transcribe(self, result: Results, audio_file: Path) -> str:
-        return ''
+    def is_loaded(self) -> bool:
+        pass
 
     @abstractmethod
-    def on_setting(self, setting: Setting):
-        pass
+    def load(self) -> None:
+        pass
+
+    @abstractmethod
+    def recognize(self, result: Results, audio: np.ndarray) -> None:
+        pass
+
+    @abstractmethod
+    def set_config(self, config: Any) -> None:
+        pass

stt_voice_messages/models/openai_whisper.py

@@ -19,8 +19,8 @@ from dataclasses import dataclass
 from pathlib import Path
 
 from ..helper import Results
-from ..model_settings import OpenAIWhisperSettings
-from .model import Model
+from .model_settings import OpenAIWhisperSettings
+from .model_template import Model
 
 log = logging.getLogger('gajim.p.sttvm_whisper')
 
@@ -37,7 +37,6 @@ class Configuration:
 
 class WhisperModel(Model):
     def __init__(self):
-        # TODO
         self._result: str = ''
         self._config = OpenAIWhisperSettings()
 
@@ -45,12 +44,11 @@ class WhisperModel(Model):
     def result(self) -> str:
         return self._result
 
-    def transcribe(self, result: Results, audio_file: Path) -> str:
-        model = whisper.load_model(self._config['whisperai_model_size'])
-        log.debug('model size is used = %s', self._config['whisperai_model_size'])
-        result.text = model.transcribe(audio_file)['text']
+    def transcribe(self, result: Results, audio_file: Path) -> None:
+        model = whisper.load_model(self._config.model_size)
+        log.debug('model size is used = %s', self._config.model_size)
+        result.text = model.transcribe(audio_file)['text']  # pyright: ignore [reportAttributeAccessIssue]
 
-    def on_setting(self, key, value):
-        log.debug('key = %s, value = %s', key, value)
-        self._config[key] = value
+    def set_config(self, config: OpenAIWhisperSettings) -> None:
+        self._config = config
 

stt_voice_messages/models/stt.py

@@ -0,0 +1,132 @@
+# This file is part of Gajim.
+#
+# Gajim is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Gajim is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Gajim. If not, see <http://www.gnu.org/licenses/>.
+
+import logging
+import pickle
+import subprocess
+import sys
+from pathlib import Path
+
+import numpy as np
+from gi.repository import GLib
+
+from ..helper import Results
+from .model_settings import OnnxAsrSettings
+from .model_template import Model
+
+log = logging.getLogger('gajim.p.sttvm_onnx_asr')
+
+
+_IDLE_UNLOAD_SECONDS = 300
+
+class OnnxAsrModel(Model):
+    def __init__(self):
+        self._proc = None
+        self._loaded = False
+        self._config = OnnxAsrSettings()
+        self._unload_source = None
+
+    @property
+    def is_loaded(self) -> bool:
+        return self._loaded
+
+    @property
+    def will_download(self) -> bool:
+        if self.is_loaded or self._config.model_path:
+            return False
+        from huggingface_hub import try_to_load_from_cache
+        from onnx_asr.resolver import model_repos
+        repo = model_repos.get(self._config.model_id, self._config.model_id)
+        if '/' not in repo:
+            return False
+        return not isinstance(try_to_load_from_cache(repo, 'config.json'), str)
+
+    def load(self) -> None:
+        if self._loaded:
+            self._reschedule_unload()
+            return
+        log.debug('Loading model %s in worker', self._config.model_id)
+        self._send({
+            'op': 'load',
+            'model_id': self._config.model_id,
+            'model_path': self._config.model_path,
+        })
+        self._loaded = True
+        self._reschedule_unload()
+
+    def recognize(self, result: Results, audio: np.ndarray) -> None:
+        self.load()
+        response = self._send({'op': 'recognize', 'audio': audio})
+        result.text = response['text']
+        self._reschedule_unload()
+
+    def set_config(self, config: OnnxAsrSettings) -> None:
+        if (config.model_id != self._config.model_id
+                or config.model_path != self._config.model_path):
+            self.unload_now()
+        self._config = OnnxAsrSettings(
+            model_id=config.model_id, model_path=config.model_path)
+
+    def unload_now(self) -> None:
+        if self._unload_source is not None:
+            GLib.source_remove(self._unload_source)
+            self._unload_source = None
+        if self._proc is not None:
+            log.debug('Terminating STT worker subprocess')
+            try:
+                self._proc.stdin.close()
+                self._proc.wait(timeout=2)
+            except subprocess.TimeoutExpired:
+                self._proc.kill()
+                self._proc.wait()
+            self._proc = None
+        self._loaded = False
+
+    def _ensure_proc(self) -> None:
+        if self._proc is not None and self._proc.poll() is None:
+            return
+        log.debug('Starting STT worker subprocess')
+        self._proc = subprocess.Popen(
+            [sys.executable, str(Path(__file__).parent / 'stt_worker.py')],
+            stdin=subprocess.PIPE,
+            stdout=subprocess.PIPE,
+        )
+        self._loaded = False
+
+    def _send(self, cmd: dict) -> dict:
+        self._ensure_proc()
+        pickle.dump(cmd, self._proc.stdin)
+        self._proc.stdin.flush()
+        try:
+            response = pickle.load(self._proc.stdout)
+        except EOFError as e:
+            self._proc = None
+            self._loaded = False
+            raise RuntimeError('Worker subprocess exited unexpectedly') from e
+        if not response.get('ok'):
+            raise RuntimeError(response.get('error', 'unknown worker error'))
+        return response
+
+    def _reschedule_unload(self) -> None:
+        if self._unload_source is not None:
+            GLib.source_remove(self._unload_source)
+        self._unload_source = GLib.timeout_add_seconds(
+            _IDLE_UNLOAD_SECONDS, self._on_idle_unload)
+
+    def _on_idle_unload(self) -> bool:
+        self._unload_source = None
+        log.debug('Idle unload after %ds', _IDLE_UNLOAD_SECONDS)
+        self.unload_now()
+        return GLib.SOURCE_REMOVE

stt_voice_messages/models/stt_worker.py

@@ -0,0 +1,54 @@
+# This file is part of Gajim.
+#
+# Gajim is free software: you can redistribute it and/or modify
+# it under the terms of the GNU General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Gajim is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Gajim. If not, see <http://www.gnu.org/licenses/>.
+
+import pickle
+import sys
+import traceback
+
+
+def _respond(response: dict) -> None:
+    pickle.dump(response, sys.stdout.buffer)
+    sys.stdout.buffer.flush()
+
+
+def main() -> None:
+    model = None
+    while True:
+        try:
+            cmd = pickle.load(sys.stdin.buffer)
+        except EOFError:
+            return
+        try:
+            op = cmd['op']
+            if op == 'load':
+                import onnx_asr
+                model = onnx_asr.load_model(
+                    cmd['model_id'], cmd.get('model_path') or None)
+                _respond({'ok': True})
+            elif op == 'recognize':
+                text = model.recognize(cmd['audio'])
+                _respond({'ok': True, 'text': text})
+            else:
+                _respond({'ok': False, 'error': f'unknown op: {op}'})
+        except Exception as e:
+            _respond({
+                'ok': False,
+                'error': f'{type(e).__name__}: {e}',
+                'traceback': traceback.format_exc(),
+            })
+
+
+if __name__ == '__main__':
+    main()

stt_voice_messages/plugin-manifest.json

@@ -13,7 +13,7 @@
         "win32"
     ],
     "requirements": [
-        "gajim>=1.9.0"
+        "gajim>=2.0.0"
     ],
     "short_name": "stt_voice_messages",
     "version": "0.0.1"

stt_voice_messages/stt_voice_messages.py

@@ -15,51 +15,89 @@
 
 from __future__ import annotations
 
+import logging
 from functools import partial
+from pathlib import Path
 
+from gi.repository import GLib, Gtk
+
+from gajim.common import app
 from gajim.plugins import GajimPlugin
 from gajim.plugins.plugins_i18n import _
 
-from .gtk.config_dialog import *
+from .gtk.config_dialog import Configuration, STTVoiceMessagesConfigDialog
 from .gtk.sttbox import STTBox
+from .models.model_settings import OnnxAsrSettings
 
 log = logging.getLogger('gajim.p.stt_voice_messages')
 
+_FOCUS_LOSS_UNLOAD_SECONDS = 30
+
 
 class STTVoiceMessagesPlugin(GajimPlugin):
     def init(self) -> None:
         self.description = _('Transcribes voice messages to text.')
+
+        self.config_default_values = {
+            'auto_transcribe': (False, ''),
+            'model_id': ('nemo-parakeet-tdt-0.6b-v3', ''),
+            'model_path': ('', ''),
+        }
+
         self._config = Configuration(self)
-        self._config.check_available_moduls()
         self.config_dialog = partial(STTVoiceMessagesConfigDialog, self._config)
 
         self.gui_extension_points = {
             'preview_audio': (self._on_preview_audio_created, None),
         }
 
-        self.config_default_values = {
-            'auto_transcribe': (False, ''),
-            'model': ('', ''),
-            'model_class': (None, ''),
-            'whisperai_model_size': ('tiny', ''),
-            'whisperai_translate': (False, ''),
-        }
+        self._active_handler_id = 0
+        self._focus_unload_source = None
 
-        self._audio_file: str = ''
-        self._preview_audio_widget = None
-        self._stt_box = None
+    def activate(self) -> None:
+        if app.window is not None and self._active_handler_id == 0:
+            self._active_handler_id = app.window.connect(
+                'notify::is-active', self._on_window_active_changed)
+
+    def deactivate(self) -> None:
+        if self._focus_unload_source is not None:
+            GLib.source_remove(self._focus_unload_source)
+            self._focus_unload_source = None
+        if self._active_handler_id != 0 and app.window is not None:
+            app.window.disconnect(self._active_handler_id)
+            self._active_handler_id = 0
+        if self._config.is_available:
+            self._config.unload_model()
+
+    def _on_window_active_changed(self,
+                                  window: Gtk.Window,
+                                  _pspec: object,
+                                  ) -> None:
+        if window.is_active():
+            if self._focus_unload_source is not None:
+                GLib.source_remove(self._focus_unload_source)
+                self._focus_unload_source = None
+        elif self._focus_unload_source is None:
+            self._focus_unload_source = GLib.timeout_add_seconds(
+                _FOCUS_LOSS_UNLOAD_SECONDS, self._on_focus_unload_fired)
+
+    def _on_focus_unload_fired(self) -> bool:
+        self._focus_unload_source = None
+        if self._config.is_available:
+            self._config.unload_model()
+        return GLib.SOURCE_REMOVE
 
     def _on_preview_audio_created(self,
-                                  preview_audio_widget: Gtk.Box,
+                                  drawing_box: Gtk.Box,
+                                  control_box: Gtk.Box,
                                   audio_file: Path
                                   ) -> None:
-        self._preview_audio_widget = preview_audio_widget
+        self._drawing_box = drawing_box;
+        self._control_box = control_box;
         self._audio_file = audio_file.as_posix()
         self._create_stt_box()
 
     def _create_stt_box(self) -> None:
-        assert self._preview_audio_widget is not None
-        self._stt_box = STTBox(self._preview_audio_widget,
-                                      self.config,
-                                      self._audio_file)
-        self._preview_audio_widget.pack_end(self._stt_box, False, False, 0)
+        self._stt_box = STTBox(self._config, self._audio_file)
+        self._control_box.append(self._stt_box.button)
+        self._drawing_box.append(self._stt_box)