diff --git a/moshi/moshi/models/lm.py b/moshi/moshi/models/lm.py
index a64d736b..c3e26584 100644
--- a/moshi/moshi/models/lm.py
+++ b/moshi/moshi/models/lm.py
@@ -657,6 +657,8 @@ def __init__(
         report_loss: bool = False,
         return_logits: bool = False,
         audio_silence_frame_cnt: int = 1,
+        on_silence_start: Optional[Callable] = None,
+        on_speech_start: Optional[Callable] = None,
         text_prompt_tokens: Optional[list[int]] = None,
         save_voice_prompt_embeddings: bool = False,
         sample_rate: int = 32000,
@@ -673,6 +675,8 @@ def __init__(
         self.top_k_text = top_k_text
         self.text_prompt_tokens = text_prompt_tokens
         self.audio_silence_frame_cnt = audio_silence_frame_cnt
+        self.on_silence_start = on_silence_start
+        self.on_speech_start = on_speech_start
         self.voice_prompt = None
         self.zero_text_code = 3
         self._frame_rate = frame_rate
@@ -1089,9 +1093,15 @@ def _step_audio_silence(self):
             pass
 
     async def _step_audio_silence_async(self, is_alive: Optional[Callable]=None):
+        # Notify silence started if callback provided
+        if self.on_silence_start is not None:
+            await self.on_silence_start()
         for _ in self._step_audio_silence_core():
             if is_alive is not None and not await is_alive():
                 break
+        # Notify silence ended (speech starting) if callback provided
+        if self.on_speech_start is not None:
+            await self.on_speech_start()
 
     def _step_text_prompt_core(self) -> Iterator[None]:
         for text_prompt_token in self.text_prompt_tokens: