WIP

src/faster_whisper_server/model_manager.py

@@ -8,21 +8,19 @@
 from typing import TYPE_CHECKING
 
 from faster_whisper import WhisperModel
+from piper.voice import PiperVoice
 
 from faster_whisper_server.hf_utils import get_piper_voice_model_file
 
 if TYPE_CHECKING:
     from collections.abc import Callable
 
-    from piper.voice import PiperVoice
-
     from faster_whisper_server.config import (
         WhisperConfig,
     )
 
 logger = logging.getLogger(__name__)
 
-
 # TODO: enable concurrent model downloads
 
 
@@ -149,8 +147,6 @@ def __init__(self, ttl: int) -> None:
         self._lock = threading.Lock()
 
     def _load_fn(self, model_id: str) -> PiperVoice:
-        from piper.voice import PiperVoice
-
         model_path = get_piper_voice_model_file(model_id)
         return PiperVoice.load(model_path)
 
@@ -167,8 +163,6 @@ def unload_model(self, model_name: str) -> None:
             self.loaded_models[model_name].unload()
 
     def load_model(self, model_name: str) -> SelfDisposingModel[PiperVoice]:
-        from piper.voice import PiperVoice
-
         with self._lock:
             if model_name in self.loaded_models:
                 logger.debug(f"{model_name} model already loaded")

tests/speech_test.py

@@ -16,6 +16,8 @@
 DEFAULT_INPUT = "Hello, world!"
 
 platform_machine = platform.machine()
+if platform_machine != "x86_64":
+    pytest.skip("Only supported on x86_64", allow_module_level=True)
 
 
 @pytest.mark.asyncio