Auto trim TTS audio

src/ovtk_audiencekit/core/Audio.py

@@ -25,9 +25,11 @@ os.close(old_stderr)
 logger = logging.getLogger(__name__)
 
 class Clip:
-    def __init__(self, path, samplerate=None, speed=1, keep_pitch=True, force_stereo=True):
+    def __init__(self, path, samplerate=None, speed=1, keep_pitch=True, force_stereo=True, trim=False):
         self.path = path
         raw, native_rate = librosa.load(self.path, sr=None, dtype='float32', mono=False)
+        if trim:
+            raw, _ = librosa.effects.trim(raw)
 
         self.channels = raw.shape[0] if len(raw.shape) == 2 else 1
         if force_stereo and self.channels == 1:

src/ovtk_audiencekit/plugins/TTS/TTS.py

@@ -1,6 +1,7 @@
 import uuid
 import os
 import asyncio
+import shutil
 
 from TTS.utils.synthesizer import Synthesizer
 from TTS.utils.manage import ModelManager
@@ -29,6 +30,7 @@ class TextToSpeechPlugin(PluginBase):
         conf_overrides = {k[2:]: v for k, v in kwargs.items() if k.startswith('o_')}
 
         self.cache_dir = os.path.join(CACHE_DIR, 'tts')
+        self.cache = {}
         os.makedirs(os.path.dirname(self.cache_dir), exist_ok=True)
 
         self.cuda = cuda
@@ -64,6 +66,7 @@ class TextToSpeechPlugin(PluginBase):
     def close(self):
         for task in self.tasks:
             task.cancel()
+        shutil.rmtree(self.cache_dir)
 
     def make_tts_wav(self, text, filename=None):
         # Force punctuation (keeps the models from acting unpredictably)
@@ -71,6 +74,10 @@ class TextToSpeechPlugin(PluginBase):
         if not any([text.endswith(punc) for punc in '.!?:']):
             text += '.'
 
+        if cached := self.cache.get(hash(text)):
+            self.logger.info(f'Cache hit - {cached}')
+            return cached
+
         if filename is None:
             filename = os.path.join(self.cache_dir, f'{uuid.uuid1()}.wav')
 
@@ -81,6 +88,7 @@ class TextToSpeechPlugin(PluginBase):
             wav = self.synthesizer.tts(text)
 
         self.synthesizer.save_wav(wav, filename)
+        self.cache[hash(text)] = filename
         self.logger.info(f'Done - saved as {filename}')
         return filename
 
@@ -90,7 +98,7 @@ class TextToSpeechPlugin(PluginBase):
             filename = await asyncio.get_running_loop().run_in_executor(None, self.make_tts_wav, text)
 
             # TODO: Play direct from memory
-            clip = Clip(filename, force_stereo=True, samplerate=self.sample_rate)
+            clip = Clip(filename, force_stereo=True, samplerate=self.sample_rate, trim=True)
             stream = Stream(clip, self.output_index)
             async def play():
                 try:

src/ovtk_audiencekit/plugins/builtins/Command.py

@@ -103,14 +103,14 @@ class CommandPlugin(PluginBase):
             raise TypeError('Should be instance of Command')
         self.commands[cmd.name] = (cmd, None, True)
 
-    def run(self, name, help=None, display=False, _children=None, **kwargs):
+    def run(self, name, help=None, display=False, required_level=None, _children=None, **kwargs):
         actionnode = next((node for node in _children if node.name == 'do'), None)
         if actionnode is None:
             raise ValueError('Command defined without an action (`do` tag)')
 
         aliases = [node.args for node in _children if node.name == 'alias']
         aliases = list(itertools.chain(*aliases))
-        cmd = Command(name, help, aliases=aliases)
+        cmd = Command(name, help, aliases=aliases, required_level=required_level)
 
         for argnode in [node for node in _children if node.name == 'arg']:
             if argnode.props.get('type'):