diff --git a/src/pcm/PcmDsd.cxx b/src/pcm/PcmDsd.cxx
index 3f79b160c..173aa5786 100644
--- a/src/pcm/PcmDsd.cxx
+++ b/src/pcm/PcmDsd.cxx
@@ -71,3 +71,41 @@ PcmDsd::ToFloat(unsigned channels, ConstBuffer<uint8_t> src)
 
 	return { dest, num_samples };
 }
+
+/**
+ * Construct a 32 bit integer from four bytes.
+ */
+static constexpr inline uint32_t
+Construct32(uint8_t a, uint8_t b, uint8_t c, uint8_t d)
+{
+	return uint32_t(a) | (uint32_t(b) << 8) |
+		(uint32_t(c) << 16) | (uint32_t(d) << 24);
+}
+
+static constexpr inline uint32_t
+Dsd8To32Sample(const uint8_t *src, unsigned channels)
+{
+	return Construct32(src[0], src[channels],
+			   src[2 * channels], src[3 * channels]);
+}
+
+ConstBuffer<uint32_t>
+Dsd8To32(PcmBuffer &buffer, unsigned channels, ConstBuffer<uint8_t> _src)
+{
+	const size_t in_frames = _src.size / channels;
+	const size_t out_frames = in_frames / 4;
+	const size_t out_samples = out_frames * channels;
+
+	const uint8_t *src = _src.data;
+	uint32_t *const dest0 = buffer.GetT<uint32_t>(out_samples);
+	uint32_t *dest = dest0;
+
+	for (size_t i = 0; i < out_frames; ++i) {
+		for (size_t c = 0; c < channels; ++c)
+			*dest++ = Dsd8To32Sample(src++, channels);
+
+		src += 3 * channels;
+	}
+
+	return {dest0, out_samples};
+}
diff --git a/src/pcm/PcmDsd.hxx b/src/pcm/PcmDsd.hxx
index 3d039bc36..ba0615252 100644
--- a/src/pcm/PcmDsd.hxx
+++ b/src/pcm/PcmDsd.hxx
@@ -48,4 +48,10 @@ public:
 				   ConstBuffer<uint8_t> src);
 };
 
+/**
+ * Convert DSD_U8 to DSD_U32 (native endian).
+ */
+ConstBuffer<uint32_t>
+Dsd8To32(PcmBuffer &buffer, unsigned channels, ConstBuffer<uint8_t> src);
+
 #endif