refractoring

multilang_translator/config.py

@@ -35,3 +35,5 @@ LANG_CONFIG = {
 }
 
 os.makedirs(ANNOUNCEMENT_DIR, exist_ok=True)
+
+# TODO. use dataclasses from Multicaster with inherit

multilang_translator/text_to_speech/text_to_speech.py

@@ -9,17 +9,20 @@ from multilang_translator.utils.resample import resample_array
 from multilang_translator.text_to_speech import encode_lc3
 
 TTS_DIR = os.path.join(os.path.dirname(__file__))
+PIPER_DIR = f'{TTS_DIR}/piper'
+os.makedirs(PIPER_DIR, exist_ok=True)
 
 def synth_piper(text, model="en_US-lessac-medium",):
-    pwd = os.getcwd()
-    os.chdir(TTS_DIR)
     start = time.time()
     
     ret = subprocess.run( # TODO: wrap this whole thing in a class and open a permanent pipe to the model 
         [config.PIPER_EXE_PATH, 
          '--cuda',
+         '--data-dir', PIPER_DIR,
+         '--download-dir', PIPER_DIR,
          '--model', model, 
-         '--output-raw'], 
+         '--output-raw'
+         ], 
         input=text.encode('utf-8'),
         capture_output=True
         )
@@ -30,15 +33,13 @@ def synth_piper(text, model="en_US-lessac-medium",):
 
     log.info("Running piper for model %s took %s s", model, round(time.time() - start, 3))
 
-    with open (f'{model}.onnx.json') as f: # TODO: wrap everything into a class, store the json permanently
+    with open (f'{PIPER_DIR}/{model}.onnx.json') as f: # TODO: wrap everyth0ing into a class, store the json permanently
         model_json = json.load(f)
 
-    os.chdir(pwd)
     return model_json, audio
 
 
 # TODO: framework should probably be a dataclass that holds all the relevant informations, also model
-# TODO: make a common repo that hold the configuration dataclasses ?
 def synthesize(text, target_sample_rate, framework, model="en_US-lessac-medium", return_lc3=True):
 
     if framework == 'piper':
@@ -74,7 +75,7 @@ if __name__ == '__main__':
     )
     target_rate=16000
 
-    audio = synthesize('Hello World', target_rate, 'piper', model= 'de_DE-kerstin-low', encode_lc3=False)
+    audio = synthesize('Hello World', target_rate, 'piper', model= 'de_DE-kerstin-low', return_lc3=False)
 
     sf.write('hello.wav', audio, target_rate)
 

tests/test_system.py

@@ -1,4 +1,4 @@
-from multilang_translator.main import announcement_from_german_text
+from multilang_translator.main_local import announcement_from_german_text
 from multilang_translator.translator import test_content
 
 

tests/test_translator.py

@@ -1,6 +1,6 @@
 from multilang_translator.translator.llm_translator import translator_de_en, translator_de_fr, translator_de_it
 from multilang_translator.translator.test_content import TESTSENTENCE_DE_BROKER, TESTSENTENCE_DE_RAINBOW
-from multilang_translator.main import translate_from_german
+from multilang_translator.main_local import translate_from_german
 
 
 import time