diff --git a/esphome/components/voice_assistant/voice_assistant.cpp b/esphome/components/voice_assistant/voice_assistant.cpp
index 50add6b1d0..dba9b925d0 100644
--- a/esphome/components/voice_assistant/voice_assistant.cpp
+++ b/esphome/components/voice_assistant/voice_assistant.cpp
@@ -249,7 +249,11 @@ void VoiceAssistant::stream_api_audio_() {
     }
 
     if (!this->api_client_->send_message(msg)) {
-      ESP_LOGV(TAG, "Audio frame dropped, TCP buffer full");
+      // Keep the chunk exposed and retry next pass, the same shape as
+      // APIConnection::try_send_camera_image_(): the slice is only lost if
+      // the ring buffer overflows before the TCP buffer clears, instead of
+      // on every refusal. The api layer already reports the refusal at V.
+      return;
     }
 
     this->audio_source_->consume(available);