diff --git a/src/core/save_state_version.h b/src/core/save_state_version.h
index ae244d871..873868575 100644
--- a/src/core/save_state_version.h
+++ b/src/core/save_state_version.h
@@ -2,4 +2,4 @@
 #include "types.h"
 
 static constexpr u32 SAVE_STATE_MAGIC = 0x43435544;
-static constexpr u32 SAVE_STATE_VERSION = 6;
+static constexpr u32 SAVE_STATE_VERSION = 7;
diff --git a/src/core/spu.cpp b/src/core/spu.cpp
index 526a6f5d5..570e1e176 100644
--- a/src/core/spu.cpp
+++ b/src/core/spu.cpp
@@ -30,6 +30,8 @@ void SPU::Initialize(System* system, DMA* dma, InterruptController* interrupt_co
 
 void SPU::Reset()
 {
+  m_tick_counter = 0;
+
   m_SPUCNT.bits = 0;
   m_SPUSTAT.bits = 0;
   m_transfer_address = 0;
@@ -67,12 +69,15 @@ void SPU::Reset()
     v.has_samples = false;
   }
 
+  m_voice_key_on_off_delay.fill(0);
+
   m_ram.fill(0);
   UpdateEventInterval();
 }
 
 bool SPU::DoState(StateWrapper& sw)
 {
+  sw.Do(&m_tick_counter);
   sw.Do(&m_SPUCNT.bits);
   sw.Do(&m_SPUSTAT.bits);
   sw.Do(&m_transfer_control.bits);
@@ -110,6 +115,8 @@ bool SPU::DoState(StateWrapper& sw)
     sw.Do(&v.has_samples);
   }
 
+  sw.Do(&m_voice_key_on_off_delay);
+
   sw.DoBytes(m_ram.data(), RAM_SIZE);
 
   if (sw.IsReading())
@@ -683,6 +690,15 @@ void SPU::Execute(TickCount ticks)
     output_stream->EndWrite(frames_in_this_batch);
     remaining_frames -= frames_in_this_batch;
   }
+
+  for (u32 i = 0; i < NUM_VOICES; i++)
+  {
+    const u32 delay = static_cast<u32>(m_voice_key_on_off_delay[i]);
+    if (delay == 0)
+      continue;
+
+    m_voice_key_on_off_delay[i] -= static_cast<u8>(std::min(delay, static_cast<u32>(ticks)));
+  }
 }
 
 void SPU::UpdateEventInterval()
@@ -1113,13 +1129,31 @@ std::tuple<s32, s32> SPU::SampleVoice(u32 voice_index)
 void SPU::VoiceKeyOn(u32 voice_index)
 {
   Log_DebugPrintf("Voice %u key on", voice_index);
+
+  if (m_voice_key_on_off_delay[voice_index] > 0)
+  {
+    Log_DevPrintf("Ignoring key on for voice %u due to only %u ticks passed since last write", voice_index,
+                  m_voice_key_on_off_delay[voice_index]);
+    return;
+  }
+
   m_voices[voice_index].KeyOn();
+  m_voice_key_on_off_delay[voice_index] = MINIMUM_TICKS_BETWEEN_KEY_ON_OFF;
 }
 
 void SPU::VoiceKeyOff(u32 voice_index)
 {
   Log_DebugPrintf("Voice %u key off", voice_index);
+
+  if (m_voice_key_on_off_delay[voice_index] > 0)
+  {
+    Log_DevPrintf("Ignoring key off for voice %u due to only %u ticks passed since last write", voice_index,
+                  m_voice_key_on_off_delay[voice_index]);
+    return;
+  }
+
   m_voices[voice_index].KeyOff();
+  m_voice_key_on_off_delay[voice_index] = MINIMUM_TICKS_BETWEEN_KEY_ON_OFF;
 }
 
 void SPU::EnsureCDAudioSpace(u32 remaining_frames)
diff --git a/src/core/spu.h b/src/core/spu.h
index b86edebdb..16a267a1e 100644
--- a/src/core/spu.h
+++ b/src/core/spu.h
@@ -56,6 +56,7 @@ private:
   static constexpr s16 ADSR_MAX_VOLUME = 0x7FFF;
   static constexpr u32 CD_AUDIO_SAMPLE_BUFFER_SIZE = 44100 * 2;
   static constexpr u32 CAPTURE_BUFFER_SIZE_PER_CHANNEL = 0x400;
+  static constexpr u32 MINIMUM_TICKS_BETWEEN_KEY_ON_OFF = 2;
 
   enum class RAMTransferMode : u8
   {
@@ -283,6 +284,7 @@ private:
   DMA* m_dma = nullptr;
   InterruptController* m_interrupt_controller = nullptr;
   std::unique_ptr<TimingEvent> m_tick_event;
+  u32 m_tick_counter = 0;
 
   SPUCNT m_SPUCNT = {};
   SPUSTAT m_SPUSTAT = {};
@@ -310,6 +312,7 @@ private:
   TickCount m_ticks_carry = 0;
 
   std::array<Voice, NUM_VOICES> m_voices{};
+  std::array<u8, NUM_VOICES> m_voice_key_on_off_delay{};
   std::array<u8, RAM_SIZE> m_ram{};
 
   InlineFIFOQueue<s16, CD_AUDIO_SAMPLE_BUFFER_SIZE> m_cd_audio_buffer;