diff --git a/modules/audio_coding/neteq/decision_logic_normal.cc b/modules/audio_coding/neteq/decision_logic_normal.cc
index c78063cc6a..10f501ac9a 100644
--- a/modules/audio_coding/neteq/decision_logic_normal.cc
+++ b/modules/audio_coding/neteq/decision_logic_normal.cc
@@ -189,10 +189,9 @@ Operations DecisionLogicNormal::FuturePacketAvailable(
   // If previous was comfort noise, then no merge is needed.
   if (prev_mode == kModeRfc3389Cng ||
       prev_mode == kModeCodecInternalCng) {
-    // Keep the same delay as before the CNG (or maximum 70 ms in buffer as
-    // safety precaution), but make sure that the number of samples in buffer
-    // is no higher than 4 times the optimal level. (Note that TargetLevel()
-    // is in Q8.)
+    // Keep the same delay as before the CNG, but make sure that the number of
+    // samples in buffer is no higher than 4 times the optimal level. (Note that
+    // TargetLevel() is in Q8.)
     if (static_cast<uint32_t>(generated_noise_samples + target_timestamp) >=
             available_timestamp ||
         cur_size_samples >
diff --git a/modules/audio_coding/neteq/neteq_impl.cc b/modules/audio_coding/neteq/neteq_impl.cc
index e7757ae5b1..b7bc0dae5d 100644
--- a/modules/audio_coding/neteq/neteq_impl.cc
+++ b/modules/audio_coding/neteq/neteq_impl.cc
@@ -47,6 +47,7 @@
 #include "rtc_base/safe_conversions.h"
 #include "rtc_base/sanitizer.h"
 #include "rtc_base/trace_event.h"
+#include "system_wrappers/include/field_trial.h"
 
 namespace webrtc {
 
@@ -102,7 +103,9 @@ NetEqImpl::NetEqImpl(const NetEq::Config& config,
       playout_mode_(config.playout_mode),
       enable_fast_accelerate_(config.enable_fast_accelerate),
       nack_enabled_(false),
-      enable_muted_state_(config.enable_muted_state) {
+      enable_muted_state_(config.enable_muted_state),
+      use_dtx_delay_fix_(
+          field_trial::IsEnabled("WebRTC-NetEqOpusDtxDelayFix")) {
   LOG(LS_INFO) << "NetEq config: " << config.ToString();
   int fs = config.sample_rate_hz;
   if (fs != 8000 && fs != 16000 && fs != 32000 && fs != 48000) {
@@ -862,6 +865,7 @@ int NetEqImpl::GetAudioInternal(AudioFrame* audio_frame, bool* muted) {
 
   AudioDecoder::SpeechType speech_type;
   int length = 0;
+  const size_t start_num_packets = packet_list.size();
   int decode_return_value = Decode(&packet_list, &operation,
                                    &length, &speech_type);
 
@@ -871,7 +875,14 @@ int NetEqImpl::GetAudioInternal(AudioFrame* audio_frame, bool* muted) {
   vad_->Update(decoded_buffer_.get(), static_cast<size_t>(length), speech_type,
                sid_frame_available, fs_hz_);
 
-  if (sid_frame_available || speech_type == AudioDecoder::kComfortNoise) {
+  // This is the criterion that we did decode some data through the speech
+  // decoder, and the operation resulted in comfort noise.
+  const bool codec_internal_sid_frame =
+      use_dtx_delay_fix_ ? (speech_type == AudioDecoder::kComfortNoise &&
+                            start_num_packets > packet_list.size())
+                         : (speech_type == AudioDecoder::kComfortNoise);
+
+  if (sid_frame_available || codec_internal_sid_frame) {
     // Start a new stopwatch since we are decoding a new CNG packet.
     generated_noise_stopwatch_ = tick_timer_->GetNewStopwatch();
   }
diff --git a/modules/audio_coding/neteq/neteq_impl.h b/modules/audio_coding/neteq/neteq_impl.h
index be986fb07e..8dca6a5fbd 100644
--- a/modules/audio_coding/neteq/neteq_impl.h
+++ b/modules/audio_coding/neteq/neteq_impl.h
@@ -440,6 +440,7 @@ class NetEqImpl : public webrtc::NetEq {
   std::unique_ptr<TickTimer::Stopwatch> generated_noise_stopwatch_
       RTC_GUARDED_BY(crit_sect_);
   std::vector<uint32_t> last_decoded_timestamps_ RTC_GUARDED_BY(crit_sect_);
+  const bool use_dtx_delay_fix_ RTC_GUARDED_BY(crit_sect_);
 
  private:
   RTC_DISALLOW_COPY_AND_ASSIGN(NetEqImpl);