feat: kick off ASR engine support for Apple Silicon

.gitignore

@@ -22,8 +22,6 @@ debug.log
 .env
 .last-skill-npm-install
 leon.json
-bridges/python/src/Pipfile.lock
-tcp_server/src/Pipfile.lock
 tcp_server/src/lib/tts/models/*.pth
 tcp_server/src/lib/tts/models/**/*.bin
 tcp_server/src/lib/tts/models/**/*.json

scripts/setup/setup-python-dev-env.js

@@ -204,7 +204,7 @@ SPACY_MODELS.set('fr', {
     }
   }
   const installPythonPackages = async () => {
-    LogHelper.info(`Installing Python packages from ${pipfilePath}.lock...`)
+    LogHelper.info(`Installing Python packages from ${pipfilePath}...`)
 
     // Delete .venv directory to reset the development environment
     if (hasDotVenv) {
@@ -214,7 +214,7 @@ SPACY_MODELS.set('fr', {
     }
 
     try {
-      await command('pipenv install --verbose --site-packages', {
+      await command('pipenv install --verbose --skip-lock', {
         shell: true,
         stdio: 'inherit'
       })

tcp_server/src/lib/asr/api.py

@@ -6,7 +6,7 @@ import numpy as np
 from faster_whisper import WhisperModel
 
 from ..constants import ASR_MODEL_PATH_FOR_GPU, ASR_MODEL_PATH_FOR_CPU
-from ..utils import ThrottledCallback
+from ..utils import ThrottledCallback, is_macos
 
 
 class ASR:
@@ -20,21 +20,21 @@ class ASR:
         self.log('Loading model...')
 
         if device == 'auto':
-            device = 'cpu'
 
             if torch.cuda.is_available(): device = 'cuda'
             else: self.log('GPU not available. CUDA is not installed?')
 
-            if torch.backends.mps.is_available(): device = 'mps'
         if 'cuda' in device:
             assert torch.cuda.is_available()
 
         self.log(f'Device: {device}')
 
-        compute_type = "float16"
+        compute_type = 'float16'
+        if is_macos():
+            compute_type = 'int8_float32'
 
         if device == 'cpu':
-            compute_type = "int8_float32"
+            compute_type = 'int8_float32'
 
         self.compute_type = compute_type
 
@@ -169,7 +169,7 @@ class ASR:
         self.log("Recording...")
         frames = []
         while True:
-            data = self.stream.read(self.chunk)
+            data = self.stream.read(self.chunk, exception_on_overflow=False)
             data_np = np.frombuffer(data, dtype=np.int16)
 
             # Check if the audio data contains any non-finite values

tcp_server/src/lib/utils.py

@@ -1,4 +1,5 @@
 import time
+import sys
 
 
 class ThrottledCallback:
@@ -12,3 +13,15 @@ class ThrottledCallback:
         if current_time - self.last_call > self.min_interval:
             self.callback(*args, **kwargs)
             self.last_call = current_time
+
+
+def is_macos():
+    return sys.platform == 'darwin'
+
+
+def is_windows():
+    return sys.platform == 'win32'
+
+
+def is_linux():
+    return sys.platform == 'linux'