diff --git a/core/transcriber.cpp b/core/transcriber.cpp
index 77c2733..e572335 100644
--- a/core/transcriber.cpp
+++ b/core/transcriber.cpp
@@ -426,10 +426,27 @@ void Transcriber::transcribe_stream(int32_t stream_id, uint32_t flags,
     std::lock_guard<std::mutex> lock(stream->vad_mutex);
     stream->vad->process_audio(audio_data, (int32_t)audio_length,
                                INTERNAL_SAMPLE_RATE);
-    segments = *(stream->vad->get_segments());
+    const std::vector<VoiceActivitySegment> *vad_segments =
+        stream->vad->get_segments();
+    segments.reserve(vad_segments->size());
+    for (const VoiceActivitySegment &segment : *vad_segments) {
+      VoiceActivitySegment segment_copy;
+      segment_copy.start_time = segment.start_time;
+      segment_copy.end_time = segment.end_time;
+      segment_copy.is_complete = segment.is_complete;
+      segment_copy.just_updated = segment.just_updated;
+      if (segment.just_updated) {
+        segment_copy.audio_data = segment.audio_data;
+      }
+      segments.push_back(std::move(segment_copy));
+    }
   }
   stream->clear_new_audio_buffer();
   this->update_transcript_from_segments(segments, stream, out_transcript);
+  if (!this->options.return_audio_data) {
+    std::lock_guard<std::mutex> lock(stream->vad_mutex);
+    stream->vad->clear_completed_segment_audio_data();
+  }
 }

 std::string Transcriber::transcript_to_string(
diff --git a/core/voice-activity-detector.cpp b/core/voice-activity-detector.cpp
index f49fbcc..58fa083 100644
--- a/core/voice-activity-detector.cpp
+++ b/core/voice-activity-detector.cpp
@@ -96,6 +96,14 @@ void VoiceActivityDetector::process_audio(const float *audio_data,
   processing_remainder_audio_buffer = processing_buffer;
 }

+void VoiceActivityDetector::clear_completed_segment_audio_data() {
+  for (VoiceActivitySegment &segment : segments) {
+    if (segment.is_complete && !segment.audio_data.empty()) {
+      std::vector<float>().swap(segment.audio_data);
+    }
+  }
+}
+
 void VoiceActivityDetector::process_audio_chunk(const float *audio_data,
                                                 size_t audio_data_size) {
   assert(audio_data_size == (size_t)(hop_size));
diff --git a/core/voice-activity-detector.h b/core/voice-activity-detector.h
index 3279bc8..bfee186 100644
--- a/core/voice-activity-detector.h
+++ b/core/voice-activity-detector.h
@@ -56,6 +56,7 @@ class VoiceActivityDetector {
   const std::vector<VoiceActivitySegment> *get_segments() const {
     return &segments;
   }
+  void clear_completed_segment_audio_data();
   std::string to_string() const;

  private:
@@ -66,4 +67,4 @@ class VoiceActivityDetector {
   void process_audio_chunk(const float *audio_data, size_t audio_data_size);
 };

-#endif
\ No newline at end of file
+#endif