diff --git a/funasr/models/fsmn_vad_streaming/model.py b/funasr/models/fsmn_vad_streaming/model.py
index f62e36a2c..7dfb96a05 100644
--- a/funasr/models/fsmn_vad_streaming/model.py
+++ b/funasr/models/fsmn_vad_streaming/model.py
@@ -724,7 +724,7 @@ def inference(
             if len(segments_i) > 0:
                 segments.extend(*segments_i)
 
-        cache["prev_samples"] = audio_sample[:-m]
+        cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
         if _is_final:
             self.init_cache(cache)
 
diff --git a/funasr/models/paraformer_streaming/model.py b/funasr/models/paraformer_streaming/model.py
index 16021ceb6..cbf666992 100644
--- a/funasr/models/paraformer_streaming/model.py
+++ b/funasr/models/paraformer_streaming/model.py
@@ -642,7 +642,7 @@ def inference(
         result_i = {"key": key[0], "text": text_postprocessed}
         result = [result_i]
 
-        cache["prev_samples"] = audio_sample[:-m]
+        cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
         if _is_final:
             self.init_cache(cache, **kwargs)
 
diff --git a/funasr/models/scama/model.py b/funasr/models/scama/model.py
index c15f435a4..22a48bccc 100644
--- a/funasr/models/scama/model.py
+++ b/funasr/models/scama/model.py
@@ -637,7 +637,7 @@ def init_cache(self, cache: dict = {}, **kwargs):
         cache["decoder"] = cache_decoder
         cache["frontend"] = {}
 
-        cache["prev_samples"] = torch.empty(0).to(device=device)
+        cache["prev_samples"] = torch.empty(0)
 
         return cache
 
@@ -726,7 +726,7 @@ def inference(
         result_i = {"key": key[0], "text": text_postprocessed}
         result = [result_i]
 
-        cache["prev_samples"] = audio_sample[:-m]
+        cache["prev_samples"] = audio_sample[-m:] if m > 0 else torch.empty(0)
         if _is_final:
             self.init_cache(cache, **kwargs)