diff --git a/modules/audio_coding/codecs/opus/audio_encoder_opus_unittest.cc b/modules/audio_coding/codecs/opus/audio_encoder_opus_unittest.cc
index 6412be5f72..f1953eaacf 100644
--- a/modules/audio_coding/codecs/opus/audio_encoder_opus_unittest.cc
+++ b/modules/audio_coding/codecs/opus/audio_encoder_opus_unittest.cc
@@ -831,6 +831,7 @@ TEST(AudioEncoderOpusTest, OpusDtxFilteringHighEnergyRefreshPackets) {
   constexpr size_t kSilenceDurationSamples = kSampleRateHz * 0.2f;
   std::array<int16_t, kSilenceDurationSamples> silence;
   uint32_t rtp_timestamp = 0;
+  bool last_packet_dtx_frame = false;
   bool opus_entered_dtx = false;
   bool silence_filled = false;
   size_t timestamp_start_silence = 0;
@@ -850,10 +851,13 @@ TEST(AudioEncoderOpusTest, OpusDtxFilteringHighEnergyRefreshPackets) {
       }
       rtp_timestamp += kInputBlockSizeSamples;
     }
-    if (info.encoded_bytes < 2 && !opus_entered_dtx) {
+    EXPECT_TRUE(info.encoded_bytes > 0 || last_packet_dtx_frame);
+    last_packet_dtx_frame = info.encoded_bytes > 0 ? info.encoded_bytes <= 2
+                                                   : last_packet_dtx_frame;
+    if (info.encoded_bytes <= 2 && !opus_entered_dtx) {
       timestamp_start_silence = rtp_timestamp;
     }
-    opus_entered_dtx = info.encoded_bytes < 2;
+    opus_entered_dtx = info.encoded_bytes <= 2;
   }
 
   EXPECT_TRUE(silence_filled);
@@ -880,6 +884,9 @@ TEST(AudioEncoderOpusTest, OpusDtxFilteringHighEnergyRefreshPackets) {
         info = encoder->Encode(rtp_timestamp, silence_frame, &encoded);
         rtp_timestamp += kInputBlockSizeSamples;
       }
+      EXPECT_TRUE(info.encoded_bytes > 0 || last_packet_dtx_frame);
+      last_packet_dtx_frame = info.encoded_bytes > 0 ? info.encoded_bytes <= 2
+                                                     : last_packet_dtx_frame;
       // Tracking the number of non empty packets.
       if (increase_noise && info.encoded_bytes > 2) {
         number_non_empty_packets_during_increase++;
diff --git a/modules/audio_coding/codecs/opus/opus_interface.cc b/modules/audio_coding/codecs/opus/opus_interface.cc
index 95c3bb9db1..f684452ad5 100644
--- a/modules/audio_coding/codecs/opus/opus_interface.cc
+++ b/modules/audio_coding/codecs/opus/opus_interface.cc
@@ -247,9 +247,15 @@ int WebRtcOpus_Encode(OpusEncInst* inst,
           inst, rtc::MakeArrayView(audio_in, samples),
           rtc::MakeArrayView(encoded, res))) {
     // This packet is a high energy refresh DTX packet. For avoiding an increase
-    // of the energy in the DTX region at the decoder, this packet is dropped.
-    inst->in_dtx_mode = 0;
-    return 0;
+    // of the energy in the DTX region at the decoder, this packet is
+    // substituted by a TOC byte with one empty frame.
+    // The number of frames described in the TOC byte
+    // (https://tools.ietf.org/html/rfc6716#section-3.1) are overwritten to
+    // always indicate one frame (last two bits equal to 0).
+    encoded[0] = encoded[0] & 0b11111100;
+    inst->in_dtx_mode = 1;
+    // The payload is just the TOC byte and has 1 byte as length.
+    return 1;
   }
   inst->in_dtx_mode = 0;
   return res;