Update to opus-HEAD-66611f1.

src/analysis.c

 /* Copyright (c) 2011 Xiph.Org Foundation
    Written by Jean-Marc Valin */
 /*
    Redistribution and use in source and binary forms, with or without
    modification, are permitted provided that the following conditions
    are met:
 
    - Redistributions of source code must retain the above copyright
    notice, this list of conditions and the following disclaimer.
 
@@ skipping 171 matching lines @@
    for (i=0;i<DETECT_SIZE-curr_lookahead;i++)
       psum += tonal->pmusic[i];
    for (;i<DETECT_SIZE;i++)
       psum += tonal->pspeech[i];
    psum = psum*tonal->music_confidence + (1-psum)*tonal->speech_confidence;
    /*printf("%f %f %f\n", psum, info_out->music_prob, info_out->tonality);*/
 
    info_out->music_prob = psum;
 }
 
-void tonality_analysis(TonalityAnalysisState *tonal, AnalysisInfo *info_out, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix)
+static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt_mode, const void *x, int len, int offset, int c1, int c2, int C, int lsb_depth, downmix_func downmix)
 {
     int i, b;
     const kiss_fft_state *kfft;
     VARDECL(kiss_fft_cpx, in);
     VARDECL(kiss_fft_cpx, out);
     int N = 480, N2=240;
     float * OPUS_RESTRICT A = tonal->angle;
     float * OPUS_RESTRICT dA = tonal->d_angle;
     float * OPUS_RESTRICT d2A = tonal->d2_angle;
     VARDECL(float, tonality);
@@ skipping 53 matching lines @@
        in[i].r = (kiss_fft_scalar)(w*tonal->inmem[i]);
        in[i].i = (kiss_fft_scalar)(w*tonal->inmem[N2+i]);
        in[N-i-1].r = (kiss_fft_scalar)(w*tonal->inmem[N-i-1]);
        in[N-i-1].i = (kiss_fft_scalar)(w*tonal->inmem[N+N2-i-1]);
     }
     OPUS_MOVE(tonal->inmem, tonal->inmem+ANALYSIS_BUF_SIZE-240, 240);
     remaining = len - (ANALYSIS_BUF_SIZE-tonal->mem_fill);
     downmix(x, &tonal->inmem[240], remaining, offset+ANALYSIS_BUF_SIZE-tonal->mem_fill, c1, c2, C);
     tonal->mem_fill = 240 + remaining;
     opus_fft(kfft, in, out);
+#ifndef FIXED_POINT
+    /* If there's any NaN on the input, the entire output will be NaN, so we only need to check one value. */
+    if (celt_isnan(out[0].r))
+    {
+       info->valid = 0;
+       RESTORE_STACK;
+       return;
+    }
+#endif
 
     for (i=1;i<N2;i++)
     {
        float X1r, X2r, X1i, X2i;
        float angle, d_angle, d2_angle;
        float angle2, d_angle2, d2_angle2;
        float mod1, mod2, avg_mod;
        X1r = (float)out[i].r+out[N-i].r;
        X1i = (float)out[i].i-out[N-i].i;
        X2r = (float)out[i].i+out[N-i].i;
@@ skipping 51 matching lines @@
           float binE = out[i].r*(float)out[i].r + out[N-i].r*(float)out[N-i].r
                      + out[i].i*(float)out[i].i + out[N-i].i*(float)out[N-i].i;
 #ifdef FIXED_POINT
           /* FIXME: It's probably best to change the BFCC filter initial state instead */
           binE *= 5.55e-17f;
 #endif
           E += binE;
           tE += binE*tonality[i];
           nE += binE*2.f*(.5f-noisiness[i]);
        }
+#ifndef FIXED_POINT
+       /* Check for extreme band energies that could cause NaNs later. */
+       if (!(E<1e9f) || celt_isnan(E))
+       {
+          info->valid = 0;
+          RESTORE_STACK;
+          return;
+       }
+#endif
+
        tonal->E[tonal->E_count][b] = E;
        frame_noisiness += nE/(1e-15f+E);
 
        frame_loudness += (float)sqrt(E+1e-10f);
        logE[b] = (float)log(E+1e-10f);
        tonal->lowE[b] = MIN32(logE[b], tonal->lowE[b]+.01f);
        tonal->highE[b] = MAX32(logE[b], tonal->highE[b]-.1f);
        if (tonal->highE[b] < tonal->lowE[b]+1.f)
        {
           tonal->highE[b]+=.5f;
@@ skipping 257 matching lines @@
     info->music_prob = 0;
 #endif
     /*for (i=0;i<25;i++)
        printf("%f ", features[i]);
     printf("\n");*/
 
     info->bandwidth = bandwidth;
     /*printf("%d %d\n", info->bandwidth, info->opus_bandwidth);*/
     info->noisiness = frame_noisiness;
     info->valid = 1;
-    if (info_out!=NULL)
-       OPUS_COPY(info_out, info, 1);
     RESTORE_STACK;
 }
 
 void run_analysis(TonalityAnalysisState *analysis, const CELTMode *celt_mode, const void *analysis_pcm,
                  int analysis_frame_size, int frame_size, int c1, int c2, int C, opus_int32 Fs,
                  int lsb_depth, downmix_func downmix, AnalysisInfo *analysis_info)
 {
    int offset;
    int pcm_len;
 
    if (analysis_pcm != NULL)
    {
       /* Avoid overflow/wrap-around of the analysis buffer */
       analysis_frame_size = IMIN((DETECT_SIZE-5)*Fs/100, analysis_frame_size);
 
       pcm_len = analysis_frame_size - analysis->analysis_offset;
       offset = analysis->analysis_offset;
       do {
-         tonality_analysis(analysis, NULL, celt_mode, analysis_pcm, IMIN(480, pcm_len), offset, c1, c2, C, lsb_depth, downmix);
+         tonality_analysis(analysis, celt_mode, analysis_pcm, IMIN(480, pcm_len), offset, c1, c2, C, lsb_depth, downmix);
          offset += 480;
          pcm_len -= 480;
       } while (pcm_len>0);
       analysis->analysis_offset = analysis_frame_size;
 
       analysis->analysis_offset -= frame_size;
    }
 
    analysis_info->valid = 0;
    tonality_get_info(analysis, analysis_info, frame_size);
 }