From a829fce89385ee38c03d29fdeac538e48f17b401 Mon Sep 17 00:00:00 2001
From: JarbasAi <jarbasai@mailfence.com>
Date: Fri, 6 Oct 2023 20:22:18 +0100
Subject: [PATCH] feat/dialog_transformers

---
 ovos_audio/service.py      | 10 +++++
 ovos_audio/transformers.py | 75 ++++++++++++++++++++++++++++++++++++++
 2 files changed, 85 insertions(+)
 create mode 100644 ovos_audio/transformers.py

diff --git a/ovos_audio/service.py b/ovos_audio/service.py
index 65ceb3f..2ed3423 100644
--- a/ovos_audio/service.py
+++ b/ovos_audio/service.py
@@ -5,6 +5,7 @@
 from ovos_audio.audio import AudioService
 from ovos_audio.playback import PlaybackThread
 from ovos_audio.tts import TTSFactory
+from ovos_audio.transformers import DialogTransformersService
 from ovos_audio.utils import report_timing, validate_message_context
 from ovos_bus_client import Message, MessageBusClient
 from ovos_bus_client.session import SessionManager
@@ -73,6 +74,7 @@ def __init__(self, ready_hook=on_ready, error_hook=on_error,
         self.bus = bus
         self.status.bind(self.bus)
         self.init_messagebus()
+        self.dialog_transform = DialogTransformersService(self.bus)
 
         try:
             self._maybe_reload_tts()
@@ -269,6 +271,14 @@ def handle_speak(self, message):
         stopwatch.start()
 
         utterance = message.data['utterance']
+
+        # allow dialog transformers to rewrite speech
+        utt2 = self.dialog_transform.transform(utterance, sess)
+        if utterance != utt22:
+            LOG.debug(f"original dialog: {utterance}")
+            LOG.info(f"dialog transformed to: {utt2}")
+            utterance = utt2
+
         listen = message.data.get('expect_response', False)
         self.execute_tts(utterance, sess.session_id, listen, message)
 
diff --git a/ovos_audio/transformers.py b/ovos_audio/transformers.py
new file mode 100644
index 0000000..c6fb9dc
--- /dev/null
+++ b/ovos_audio/transformers.py
@@ -0,0 +1,75 @@
+from ovos_plugin_manager.dialog_transformers import find_dialog_transformer_plugins
+from ovos_utils.json_helper import merge_dict
+from ovos_utils.log import LOG
+from ovos_bus_client.session import Session, SessionManager
+
+
+class DialogTransformersService:
+    """ transform dialogs before being sent to TTS """
+    def __init__(self, bus, config=None):
+        self.loaded_plugins = {}
+        self.has_loaded = False
+        self.bus = bus
+        # to activate a plugin, just add an entry to mycroft.conf for it
+        self.config = config or Configuration().get("dialog_transformers", {})
+        self.load_plugins()
+
+    def load_plugins(self):
+        for plug_name, plug in find_dialog_transformer_plugins().items():
+            if plug_name in self.config:
+                # if disabled skip it
+                if not self.config[plug_name].get("active", True):
+                    continue
+                try:
+                    self.loaded_plugins[plug_name] = plug()
+                    self.loaded_plugins[plug_name].bind(self.bus)
+                    LOG.info(f"loaded audio transformer plugin: {plug_name}")
+                except Exception as e:
+                    LOG.exception(f"Failed to load audio transformer plugin: "
+                                  f"{plug_name}")
+        self.has_loaded = True
+
+    @property
+    def plugins(self) -> list:
+        """
+        Return loaded transformers in priority order, such that modules with a
+        higher `priority` rank are called first and changes from lower ranked
+        transformers are applied last.
+
+        A plugin of `priority` 1 will override any existing context keys and
+        will be the last to modify `audio_data`
+        """
+        return sorted(self.loaded_plugins.values(),
+                      key=lambda k: k.priority, reverse=True)
+
+    def shutdown(self):
+        """
+        Shutdown all loaded plugins
+        """
+        for module in self.plugins:
+            try:
+                module.shutdown()
+            except Exception as e:
+                LOG.warning(e)
+
+    def transform(self, dialog: str, session: Session= None) -> str:
+        """
+        Get transformed audio and context for the preceding audio
+        @param dialog: str to be spoken
+        @return: transformed dialog to be sent to TTS
+        """
+        session = session or SessionManager.get()
+
+        # TODO property not yet introduced in Session
+        # this will be set per Session/Persona
+        # active_transformers = session.dialog_transformers or self.plugins
+        active_transformers = self.plugins
+
+        for module in active_transformers:
+            try:
+                LOG.debug(f"checking dialog transformer: {module}")
+                dialog = module.transform(dialog)
+                LOG.debug(f"{module.name}: {dialog}")
+            except Exception as e:
+                LOG.exception(e)
+        return dialog