diff --git a/Core/HLE/AtracCtx.cpp b/Core/HLE/AtracCtx.cpp
index b9849baa32..4a9c8a0e2d 100644
--- a/Core/HLE/AtracCtx.cpp
+++ b/Core/HLE/AtracCtx.cpp
@@ -959,13 +959,7 @@ u32 Atrac::DecodeData(u8 *outbuf, u32 outbufPtr, u32 *SamplesNum, u32 *finish, i
 	if (off < first_.size) {
 		uint8_t *indata = BufferStart() + off;
 		int bytesConsumed = 0;
-		int outSamples = 0;
-		if (!decoder_->Decode(indata, track_.bytesPerFrame, &bytesConsumed, outputChannels_, (int16_t *)outbuf, &outSamples)) {
-			// Decode failed.
-			*SamplesNum = 0;
-			*finish = 1;
-			return ATRAC_ERROR_ALL_DATA_DECODED;
-		}
+		int outSamples = track_.SamplesPerFrame();
 		int outBytes = outSamples * outputChannels_ * sizeof(int16_t);
 		gotFrame = true;
 
@@ -978,6 +972,14 @@ u32 Atrac::DecodeData(u8 *outbuf, u32 outbufPtr, u32 *SamplesNum, u32 *finish, i
 		// If we're at the end, clamp to samples we want.  It always returns a full chunk.
 		numSamples = std::min(maxSamples, numSamples);
 
+		outSamples = numSamples;
+		if (!decoder_->Decode(indata, track_.bytesPerFrame, &bytesConsumed, outputChannels_, (int16_t *)outbuf, &outSamples)) {
+			// Decode failed.
+			*SamplesNum = 0;
+			*finish = 1;
+			return ATRAC_ERROR_ALL_DATA_DECODED;
+		}
+
 		if (packetAddr != 0 && MemBlockInfoDetailed()) {
 			char tagData[128];
 			size_t tagSize = FormatMemWriteTagAt(tagData, sizeof(tagData), "AtracDecode/", packetAddr, track_.bytesPerFrame);
diff --git a/Core/HW/Atrac3Standalone.cpp b/Core/HW/Atrac3Standalone.cpp
index 6fb5ffc540..b284be90bd 100644
--- a/Core/HW/Atrac3Standalone.cpp
+++ b/Core/HW/Atrac3Standalone.cpp
@@ -82,6 +82,10 @@ public:
 			*inbytesConsumed = result;
 		}
 		if (outSamples) {
+			// Allow capping the output samples by setting *outSamples to non-zero.
+			if (*outSamples != 0) {
+				nb_samples = std::min(*outSamples, nb_samples);
+			}
 			*outSamples = nb_samples;
 		}
 		if (nb_samples > 0) {
diff --git a/Core/HW/SimpleAudioDec.h b/Core/HW/SimpleAudioDec.h
index 34bf3b9342..40d724c565 100644
--- a/Core/HW/SimpleAudioDec.h
+++ b/Core/HW/SimpleAudioDec.h
@@ -38,6 +38,8 @@ public:
 
 	// inbytesConsumed can include skipping metadata.
 	// outSamples is in stereo samples. So you have to multiply by 4 for 16-bit stereo audio to get bytes.
+	// For Atrac3, if *outSamples != 0, it'll cap the number of samples to output. In this case, its value can only shrink.
+	// TODO: Implement that in the other decoders too, if needed.
 	virtual bool Decode(const uint8_t *inbuf, int inbytes, int *inbytesConsumed, int outputChannels, int16_t *outbuf, int *outSamples) = 0;
 	virtual bool IsOK() const = 0;