diff --git a/fireredtts2/codec/model.py b/fireredtts2/codec/model.py
index 1bf9476..a20b6e0 100644
--- a/fireredtts2/codec/model.py
+++ b/fireredtts2/codec/model.py
@@ -211,6 +211,9 @@ def __init__(self, codec: RedCodec):
     def from_pretrained(cls, conf_path: str, ckpt_path: str) -> "RedCodecInfer":
         with open(conf_path, "r") as f:
             codec = RedCodec.from_config(conf_path)
+        # support cpu
+        device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
+        ckpt = torch.load(ckpt_path, map_location=device)["generator"]
         ckpt = torch.load(ckpt_path)["generator"]
         codec.load_state_dict(ckpt)
         return cls(codec)
diff --git a/fireredtts2/fireredtts2.py b/fireredtts2/fireredtts2.py
index 4e35c38..ed7c943 100644
--- a/fireredtts2/fireredtts2.py
+++ b/fireredtts2/fireredtts2.py
@@ -58,6 +58,9 @@ def __init__(self, pretrained_dir, gen_type, device):
 
     def load_prompt_audio(self, audio_path) -> torch.Tensor:
         audio, audio_sr = torchaudio.load(audio_path)
+        # Convert to mono if multi-channel
+        if audio.shape[0] > 1:
+            audio = torch.mean(audio, dim=0, keepdim=True)
         audio16k = torchaudio.functional.resample(audio, audio_sr, 16000)
         return audio16k