diff --git a/libavcodec/aacdec_template.c b/libavcodec/aacdec_template.c
index 4b98142536..2a2d43f0a4 100644
--- a/libavcodec/aacdec_template.c
+++ b/libavcodec/aacdec_template.c
@@ -1211,7 +1211,7 @@ static av_cold int aac_decode_init(AVCodecContext *avctx)
     AAC_RENAME_32(ff_mdct_init)(&ac->mdct_small,  8, 1, 1.0 / RANGE15(128.0));
     AAC_RENAME_32(ff_mdct_init)(&ac->mdct_ltp,   11, 0, RANGE15(-2.0));
 #if !USE_FIXED
-    ret = ff_mdct15_init(&ac->mdct480, 1, 5, -1.0f);
+    ret = ff_mdct15_init(&ac->mdct480, 1, 5, -1.0f/(16*1024*960));
     if (ret < 0)
         return ret;
 #endif
@@ -2664,7 +2664,7 @@ static void imdct_and_windowing_eld(AACContext *ac, SingleChannelElement *sce)
     }
 #if !USE_FIXED
     if (n == 480)
-        ac->mdct480->imdct_half(ac->mdct480, buf, in, 1, -1.f/(16*1024*960));
+        ac->mdct480->imdct_half(ac->mdct480, buf, in, 1);
     else
 #endif
         ac->mdct.imdct_half(&ac->mdct_ld, buf, in);
diff --git a/libavcodec/mdct15.c b/libavcodec/mdct15.c
index f93881fbed..cc7d833165 100644
--- a/libavcodec/mdct15.c
+++ b/libavcodec/mdct15.c
@@ -207,7 +207,7 @@ static void mdct15(MDCT15Context *s, float *dst, const float *src, ptrdiff_t str
 }
 
 static void imdct15_half(MDCT15Context *s, float *dst, const float *src,
-                         ptrdiff_t stride, float scale)
+                         ptrdiff_t stride)
 {
     FFTComplex fft15in[15];
     FFTComplex *z = (FFTComplex *)dst;
@@ -230,16 +230,11 @@ static void imdct15_half(MDCT15Context *s, float *dst, const float *src,
 
     /* Reindex again, apply twiddles and output */
     for (i = 0; i < len8; i++) {
-        float re0, im0, re1, im1;
         const int i0 = len8 + i, i1 = len8 - i - 1;
         const int s0 = s->pfa_postreindex[i0], s1 = s->pfa_postreindex[i1];
 
-        CMUL(re0, im1, s->tmp[s1].im, s->tmp[s1].re,  s->twiddle_exptab[i1].im, s->twiddle_exptab[i1].re);
-        CMUL(re1, im0, s->tmp[s0].im, s->tmp[s0].re,  s->twiddle_exptab[i0].im, s->twiddle_exptab[i0].re);
-        z[i1].re = scale * re0;
-        z[i1].im = scale * im0;
-        z[i0].re = scale * re1;
-        z[i0].im = scale * im1;
+        CMUL(z[i1].re, z[i0].im, s->tmp[s1].im, s->tmp[s1].re,  s->twiddle_exptab[i1].im, s->twiddle_exptab[i1].re);
+        CMUL(z[i0].re, z[i1].im, s->tmp[s0].im, s->tmp[s0].re,  s->twiddle_exptab[i0].im, s->twiddle_exptab[i0].re);
     }
 }
 
diff --git a/libavcodec/mdct15.h b/libavcodec/mdct15.h
index 7d83f3ebdf..1c2149d436 100644
--- a/libavcodec/mdct15.h
+++ b/libavcodec/mdct15.h
@@ -47,7 +47,7 @@ typedef struct MDCT15Context {
 
     /* Calculate the middle half of the iMDCT */
     void (*imdct_half)(struct MDCT15Context *s, float *dst, const float *src,
-                       ptrdiff_t src_stride, float scale);
+                       ptrdiff_t stride);
 } MDCT15Context;
 
 /* Init an (i)MDCT of the length 2 * 15 * (2^N) */
diff --git a/libavcodec/opus_celt.c b/libavcodec/opus_celt.c
index 3de92f0cb2..13bfe986be 100644
--- a/libavcodec/opus_celt.c
+++ b/libavcodec/opus_celt.c
@@ -775,10 +775,9 @@ int ff_celt_decode_frame(CeltFrame *f, OpusRangeCoder *rc,
                          float **output, int channels, int frame_size,
                          int start_band,  int end_band)
 {
-    int i, j;
+    int i, j, downmix = 0;
     int consumed;           // bits of entropy consumed thus far for this frame
     MDCT15Context *imdct;
-    float imdct_scale = 1.0;
 
     if (channels != 1 && channels != 2) {
         av_log(f->avctx, AV_LOG_ERROR, "Invalid number of coded channels: %d\n",
@@ -870,7 +869,7 @@ int ff_celt_decode_frame(CeltFrame *f, OpusRangeCoder *rc,
     /* stereo -> mono downmix */
     if (f->output_channels < f->channels) {
         f->dsp->vector_fmac_scalar(f->block[0].coeffs, f->block[1].coeffs, 1.0, FFALIGN(frame_size, 16));
-        imdct_scale = 0.5;
+        downmix = 1;
     } else if (f->output_channels > f->channels)
         memcpy(f->block[1].coeffs, f->block[0].coeffs, frame_size * sizeof(float));
 
@@ -895,11 +894,14 @@ int ff_celt_decode_frame(CeltFrame *f, OpusRangeCoder *rc,
             float *dst  = block->buf + 1024 + j * f->blocksize;
 
             imdct->imdct_half(imdct, dst + CELT_OVERLAP / 2, f->block[i].coeffs + j,
-                              f->blocks, imdct_scale);
+                              f->blocks);
             f->dsp->vector_fmul_window(dst, dst, dst + CELT_OVERLAP / 2,
                                        ff_celt_window, CELT_OVERLAP / 2);
         }
 
+        if (downmix)
+            f->dsp->vector_fmul_scalar(&block->buf[1024], &block->buf[1024], 0.5f, frame_size);
+
         /* postfilter */
         celt_postfilter(f, block);