diff --git a/Core/HW/SasAudio.cpp b/Core/HW/SasAudio.cpp
index fb9254b1fe..542b3953f1 100644
--- a/Core/HW/SasAudio.cpp
+++ b/Core/HW/SasAudio.cpp
@@ -48,7 +48,7 @@ static const s8 f[16][2] = {
 	{   0,   0 },
 };
 
-void VagDecoder::Start(u32 data, int vagSize, bool loopEnabled) {
+void VagDecoder::Start(u32 data, u32 vagSize, bool loopEnabled) {
 	loopEnabled_ = loopEnabled;
 	loopAtNextBlock_ = false;
 	loopStartBlock_ = 0;
@@ -462,6 +462,9 @@ void SasInstance::MixVoice(SasVoice &voice) {
 		const bool ignorePitch = voice.type == VOICETYPE_PCM && voice.pitch > PSP_SAS_PITCH_BASE;
 		if (voice.envelope.NeedsKeyOn()) {
 			int delay = ignorePitch ? 32 : (32 * (u32)voice.pitch) >> PSP_SAS_PITCH_BASE_SHIFT;
+			// VAG seems to have an extra sample delay (not shared by PCM.)
+			if (voice.type == VOICETYPE_VAG)
+				++delay;
 			voice.ReadSamples(resampleBuffer + 2 + delay, numSamples - delay);
 		} else {
 			voice.ReadSamples(resampleBuffer + 2, numSamples);
diff --git a/Core/HW/SasAudio.h b/Core/HW/SasAudio.h
index ffcf66eddb..0c825652ab 100644
--- a/Core/HW/SasAudio.h
+++ b/Core/HW/SasAudio.h
@@ -88,7 +88,7 @@ enum VoiceType {
 class VagDecoder {
 public:
 	VagDecoder() : data_(0), read_(0), end_(true) {}
-	void Start(u32 dataPtr, int vagSize, bool loopEnabled);
+	void Start(u32 dataPtr, u32 vagSize, bool loopEnabled);
 
 	void GetSamples(s16 *outSamples, int numSamples);
 
@@ -229,7 +229,7 @@ struct SasVoice {
 	VoiceType type;
 
 	u32 vagAddr;
-	int vagSize;
+	u32 vagSize;
 	u32 pcmAddr;
 	int pcmSize;
 	int pcmIndex;