More settings tests

local_whisper/audio.py

@@ -3,7 +3,9 @@ from typing import BinaryIO
 import ffmpeg
 import numpy as np
 
-DEFAULT_SAMPLE_RATE = 16000
+from .settings import whisper_settings
+
+DEFAULT_SAMPLE_RATE = whisper_settings.default_sample_rate
 
 
 # TODO probably can offload this on a worker queue too

local_whisper/settings.py

@@ -10,6 +10,10 @@ class WhisperSettings(BaseSettings):
     base_asr_model: str = Field(
         default="medium.en", description="The base whisper model to host."
     )
+    default_sample_rate: int = Field(
+        default=16000,
+        description="The default sample rate used to resample the audio if necessary",
+    )
 
     class Config:
         env_prefix = "WHISPER_"

tests/unit/test_inference.py

@@ -8,6 +8,6 @@ def test_transcribe(sample_audio):
     result = transcribe(audio)
     assert (
         result["text"].strip().lower()
-        == "Let's see, right now I'm playing Horizon Zero Dawn. \
+        == "Let's see, right now I'm playing Horizon Zero Dawn."
-              I also had just recently finished BioShock Infinite.".lower()
+        " I also had just recently finished BioShock Infinite.".lower()
     )

tests/unit/test_settings.py

@@ -5,9 +5,7 @@ from pydantic.types import SecretStr
 
 from local_whisper.settings import WhisperSettings
 
-SETTING_DEFAULTS = {
+SETTING_DEFAULTS = {"BASE_ASR_MODEL": "medium.en", "DEFAULT_SAMPLE_RATE": 16000}
-    "BASE_ASR_MODEL": "medium.en",
-}
 
 
 def test_setting_defaults():
@@ -23,3 +21,9 @@ def test_setting_defaults():
                 else _setting_value
             )
             unmasked_setting == v
+
+
+def test_with_envvar_prefix():
+    with patch.dict(os.environ, {"WHISPER_DEFAULT_SAMPLE_RATE": "22500"}, clear=True):
+        whisper_settings = WhisperSettings()
+        assert whisper_settings.default_sample_rate == 22500