diff --git a/modules/speech_recognition/faster_whisper_module.py b/modules/speech_recognition/faster_whisper_module.py
index aedb95a..c571257 100644
--- a/modules/speech_recognition/faster_whisper_module.py
+++ b/modules/speech_recognition/faster_whisper_module.py
@@ -22,17 +22,22 @@ RECORDING_FILE_PATH = "stt_test.wav"
 model_size = "large-v3-turbo"
 
 
-def load_model(file_path=None,whisper_device="cuda",whisper_compute_type='float16'):
+def load_model(file_path=None,whisper_device="cuda",whisper_compute_type="auto"):
     """
     Load given vosk model from file or default to en-us model.
     Download model to user cache folder, example: C:/Users/toto/.cache/vosk
     """
+    if whisper_compute_type=="auto":
+        whisper_compute_type=(
+            "int8"
+            if whisper_device=="cpu"
+            else "float16")
 
     if file_path is None:
-        print(f"faster-whisper using {model_size}")
+        print(f"faster-whisper using {model_size} model with {whisper_compute_type}")
         return WhisperModel(model_size, device=whisper_device, compute_type=whisper_compute_type)
     else:
-        print(f"faster-whisper using {file_path}")
+        print(f"faster-whisper using {file_path} model with {whisper_compute_type}")
         return WhisperModel(file_path, device=whisper_device, compute_type=whisper_compute_type)
 
 def process_audio():
diff --git a/server.py b/server.py
index 38054e0..e821bd6 100644
--- a/server.py
+++ b/server.py
@@ -939,7 +939,7 @@ parser.add_argument("--stt-whisper-model-path", help="Load a custom whisper spee
 
 parser.add_argument("--use-faster-whisper", action="store_true", help="Choose to use faster-whisper instead of whisper")
 parser.add_argument("--faster-whisper-cpu", action="store_true", help="Use cpu to run faster-whisper, saves VRAM but much slower")
-parser.add_argument("--faster-whisper-type", help="Choose faster-whisper compute type, defaults to float16")
+parser.add_argument("--faster-whisper-type", help="Choose faster-whisper compute type, defaults to float16 for cuda and int8 for cpu")
 
 # sd_group = parser.add_mutually_exclusive_group()
 
@@ -1186,7 +1186,7 @@ if "whisper-stt" in modules:
         faster_whisper_type=(
             args.faster_whisper_type
             if args.faster_whisper_type
-            else "float16")
+            else "auto")
 
         print(f"Initializing Faster-Whisper speech-recognition (from ST request file) on {faster_whisper_device}")
         import modules.speech_recognition.faster_whisper_module as whisper_module