diff --git a/spa/plugins/audioconvert/fmt-ops-avx2.c b/spa/plugins/audioconvert/fmt-ops-avx2.c
index 065fa997e..5c9ea6793 100644
--- a/spa/plugins/audioconvert/fmt-ops-avx2.c
+++ b/spa/plugins/audioconvert/fmt-ops-avx2.c
@@ -41,7 +41,7 @@ conv_s16_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	const int16_t *s = src;
 	float *d0 = dst[0];
 	uint32_t n, unrolled;
-	__m256i in;
+	__m256i in = _mm256_setzero_si256();
 	__m256 out, factor = _mm256_set1_ps(1.0f / S16_SCALE);
 
 	if (SPA_LIKELY(SPA_IS_ALIGNED(d0, 32)))
@@ -67,7 +67,7 @@ conv_s16_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	}
 	for(; n < n_samples; n++) {
 		__m128 out, factor = _mm_set1_ps(1.0f / S16_SCALE);
-		out = _mm_cvtsi32_ss(out, s[0]);
+		out = _mm_cvtsi32_ss(factor, s[0]);
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += n_channels;
@@ -133,9 +133,9 @@ conv_s16_to_f32d_2_avx2(struct convert *conv, void * SPA_RESTRICT dst[], const v
 	}
 	for(; n < n_samples; n++) {
 		__m128 out[4], factor = _mm_set1_ps(1.0f / S16_SCALE);
-		out[0] = _mm_cvtsi32_ss(out[0], s[0]);
+		out[0] = _mm_cvtsi32_ss(factor, s[0]);
 		out[0] = _mm_mul_ss(out[0], factor);
-		out[1] = _mm_cvtsi32_ss(out[1], s[1]);
+		out[1] = _mm_cvtsi32_ss(factor, s[1]);
 		out[1] = _mm_mul_ss(out[1], factor);
 		_mm_store_ss(&d0[n], out[0]);
 		_mm_store_ss(&d1[n], out[1]);
@@ -175,7 +175,7 @@ conv_s24_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out = _mm_cvtsi32_ss(out, read_s24(s));
+		out = _mm_cvtsi32_ss(factor, read_s24(s));
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += 3 * n_channels;
@@ -232,8 +232,8 @@ conv_s24_to_f32d_2s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
-		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
+		out[0] = _mm_cvtsi32_ss(factor, read_s24(s));
+		out[1] = _mm_cvtsi32_ss(factor, read_s24(s+3));
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		_mm_store_ss(&d0[n], out[0]);
@@ -313,10 +313,10 @@ conv_s24_to_f32d_4s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
-		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
-		out[2] = _mm_cvtsi32_ss(out[2], read_s24(s+6));
-		out[3] = _mm_cvtsi32_ss(out[3], read_s24(s+9));
+		out[0] = _mm_cvtsi32_ss(factor, read_s24(s));
+		out[1] = _mm_cvtsi32_ss(factor, read_s24(s+3));
+		out[2] = _mm_cvtsi32_ss(factor, read_s24(s+6));
+		out[3] = _mm_cvtsi32_ss(factor, read_s24(s+9));
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		out[2] = _mm_mul_ss(out[2], factor);
@@ -406,10 +406,10 @@ conv_s32_to_f32d_4s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	}
 	for(; n < n_samples; n++) {
 		__m128 out[4], factor = _mm_set1_ps(1.0f / S24_SCALE);
-		out[0] = _mm_cvtsi32_ss(out[0], s[0]>>8);
-		out[1] = _mm_cvtsi32_ss(out[1], s[1]>>8);
-		out[2] = _mm_cvtsi32_ss(out[2], s[2]>>8);
-		out[3] = _mm_cvtsi32_ss(out[3], s[3]>>8);
+		out[0] = _mm_cvtsi32_ss(factor, s[0]>>8);
+		out[1] = _mm_cvtsi32_ss(factor, s[1]>>8);
+		out[2] = _mm_cvtsi32_ss(factor, s[2]>>8);
+		out[3] = _mm_cvtsi32_ss(factor, s[3]>>8);
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		out[2] = _mm_mul_ss(out[2], factor);
@@ -467,8 +467,8 @@ conv_s32_to_f32d_2s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	}
 	for(; n < n_samples; n++) {
 		__m128 out[2], factor = _mm_set1_ps(1.0f / S24_SCALE);
-		out[0] = _mm_cvtsi32_ss(out[0], s[0]>>8);
-		out[1] = _mm_cvtsi32_ss(out[1], s[1]>>8);
+		out[0] = _mm_cvtsi32_ss(factor, s[0]>>8);
+		out[1] = _mm_cvtsi32_ss(factor, s[1]>>8);
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		_mm_store_ss(&d0[n], out[0]);
@@ -518,7 +518,7 @@ conv_s32_to_f32d_1s_avx2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	}
 	for(; n < n_samples; n++) {
 		__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
-		out = _mm_cvtsi32_ss(out, s[0]>>8);
+		out = _mm_cvtsi32_ss(factor, s[0]>>8);
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += n_channels;
diff --git a/spa/plugins/audioconvert/fmt-ops-sse2.c b/spa/plugins/audioconvert/fmt-ops-sse2.c
index 81524d2fd..4fd13a1d4 100644
--- a/spa/plugins/audioconvert/fmt-ops-sse2.c
+++ b/spa/plugins/audioconvert/fmt-ops-sse2.c
@@ -33,7 +33,7 @@ conv_s16_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 	const int16_t *s = src;
 	float *d0 = dst[0];
 	uint32_t n, unrolled;
-	__m128i in;
+	__m128i in = _mm_setzero_si128();
 	__m128 out, factor = _mm_set1_ps(1.0f / S16_SCALE);
 
 	if (SPA_LIKELY(SPA_IS_ALIGNED(d0, 16)))
@@ -53,7 +53,7 @@ conv_s16_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 4*n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out = _mm_cvtsi32_ss(out, s[0]);
+		out = _mm_cvtsi32_ss(factor, s[0]);
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += n_channels;
@@ -118,9 +118,9 @@ conv_s16_to_f32d_2_sse2(struct convert *conv, void * SPA_RESTRICT dst[], const v
 		s += 16;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], s[0]);
+		out[0] = _mm_cvtsi32_ss(factor, s[0]);
 		out[0] = _mm_mul_ss(out[0], factor);
-		out[1] = _mm_cvtsi32_ss(out[1], s[1]);
+		out[1] = _mm_cvtsi32_ss(factor, s[1]);
 		out[1] = _mm_mul_ss(out[1], factor);
 		_mm_store_ss(&d0[n], out[0]);
 		_mm_store_ss(&d1[n], out[1]);
@@ -160,7 +160,7 @@ conv_s24_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out = _mm_cvtsi32_ss(out, read_s24(s));
+		out = _mm_cvtsi32_ss(factor, read_s24(s));
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += 3 * n_channels;
@@ -217,8 +217,8 @@ conv_s24_to_f32d_2s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
-		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
+		out[0] = _mm_cvtsi32_ss(factor, read_s24(s));
+		out[1] = _mm_cvtsi32_ss(factor, read_s24(s+3));
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		_mm_store_ss(&d0[n], out[0]);
@@ -298,10 +298,10 @@ conv_s24_to_f32d_4s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
-		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
-		out[2] = _mm_cvtsi32_ss(out[2], read_s24(s+6));
-		out[3] = _mm_cvtsi32_ss(out[3], read_s24(s+9));
+		out[0] = _mm_cvtsi32_ss(factor, read_s24(s));
+		out[1] = _mm_cvtsi32_ss(factor, read_s24(s+3));
+		out[2] = _mm_cvtsi32_ss(factor, read_s24(s+6));
+		out[3] = _mm_cvtsi32_ss(factor, read_s24(s+9));
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		out[2] = _mm_mul_ss(out[2], factor);
@@ -357,7 +357,7 @@ conv_s32_to_f32d_1s_sse2(void *data, void * SPA_RESTRICT dst[], const void * SPA
 		s += 4*n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out = _mm_cvtsi32_ss(out, s[0]>>8);
+		out = _mm_cvtsi32_ss(factor, s[0]>>8);
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += n_channels;
diff --git a/spa/plugins/audioconvert/fmt-ops-sse41.c b/spa/plugins/audioconvert/fmt-ops-sse41.c
index 0478555e8..6fb0c81cf 100644
--- a/spa/plugins/audioconvert/fmt-ops-sse41.c
+++ b/spa/plugins/audioconvert/fmt-ops-sse41.c
@@ -33,7 +33,7 @@ conv_s24_to_f32d_1s_sse41(void *data, void * SPA_RESTRICT dst[], const void * SP
 	const uint8_t *s = src;
 	float *d0 = dst[0];
 	uint32_t n, unrolled;
-	__m128i in;
+	__m128i in = _mm_setzero_si128();
 	__m128 out, factor = _mm_set1_ps(1.0f / S24_SCALE);
 
 	if (SPA_IS_ALIGNED(d0, 16))
@@ -54,7 +54,7 @@ conv_s24_to_f32d_1s_sse41(void *data, void * SPA_RESTRICT dst[], const void * SP
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out = _mm_cvtsi32_ss(out, read_s24(s));
+		out = _mm_cvtsi32_ss(factor, read_s24(s));
 		out = _mm_mul_ss(out, factor);
 		_mm_store_ss(&d0[n], out);
 		s += 3 * n_channels;
diff --git a/spa/plugins/audioconvert/fmt-ops-ssse3.c b/spa/plugins/audioconvert/fmt-ops-ssse3.c
index 6a7fc2e05..744186a60 100644
--- a/spa/plugins/audioconvert/fmt-ops-ssse3.c
+++ b/spa/plugins/audioconvert/fmt-ops-ssse3.c
@@ -77,10 +77,10 @@ conv_s24_to_f32d_4s_ssse3(void *data, void * SPA_RESTRICT dst[], const void * SP
 		s += 12 * n_channels;
 	}
 	for(; n < n_samples; n++) {
-		out[0] = _mm_cvtsi32_ss(out[0], read_s24(s));
-		out[1] = _mm_cvtsi32_ss(out[1], read_s24(s+3));
-		out[2] = _mm_cvtsi32_ss(out[2], read_s24(s+6));
-		out[3] = _mm_cvtsi32_ss(out[3], read_s24(s+9));
+		out[0] = _mm_cvtsi32_ss(factor, read_s24(s));
+		out[1] = _mm_cvtsi32_ss(factor, read_s24(s+3));
+		out[2] = _mm_cvtsi32_ss(factor, read_s24(s+6));
+		out[3] = _mm_cvtsi32_ss(factor, read_s24(s+9));
 		out[0] = _mm_mul_ss(out[0], factor);
 		out[1] = _mm_mul_ss(out[1], factor);
 		out[2] = _mm_mul_ss(out[2], factor);