shithub: opus

Download patch

ref: 90d74bbbe98de3dc979510993b01ca460b337ff3
parent: 543ee940373ab2b78494a5e502236cd8cac5f8c0
author: Jean-Marc Valin <jmvalin@jmvalin.ca>
date: Fri Feb 15 13:12:38 EST 2019 

Add bidirectional quantizer


--- a/dnn/ceps_vq_train.c
+++ b/dnn/ceps_vq_train.c
@@ -6,9 +6,12 @@
 #include <math.h>
 
 #define MIN(a,b) ((a)<(b)?(a):(b))
-#define COEF 0.75f
+#define COEF 0.0f
 #define MAX_ENTRIES 16384
 
+#define MULTI 4
+#define MULTI_MASK (MULTI-1)
+
 void compute_weights(const float *x, float *w, int ndim)
 {
   int i;
@@ -48,6 +51,31 @@
   return nearest;
 }
 
+int find_nearest_multi(const float *codebook, int nb_entries, const float *x, int ndim, float *dist)
+{
+  int i, j;
+  float min_dist = 1e15;
+  int nearest = 0;
+
+  for (i=0;i<nb_entries;i++)
+  {
+    int offset;
+    float dist=0;
+    offset = (i&MULTI_MASK)*ndim;
+    for (j=0;j<ndim;j++)
+      dist += (x[offset+j]-codebook[i*ndim+j])*(x[offset+j]-codebook[i*ndim+j]);
+    if (dist<min_dist)
+    {
+      min_dist = dist;
+      nearest = i;
+    }
+  }
+  if (dist)
+    *dist = min_dist;
+  return nearest;
+}
+
+
 int find_nearest_weighted(const float *codebook, int nb_entries, float *x, const float *w, int ndim)
 {
   int i, j;
@@ -203,6 +231,45 @@
   //fprintf(stderr, "%f / %d\n", 1./w2, nb_entries);
 }
 
+void update_multi(float *data, int nb_vectors, float *codebook, int nb_entries, int ndim)
+{
+  int i,j;
+  int count[nb_entries];
+  int nearest[nb_vectors];
+  double err=0;
+
+  for (i=0;i<nb_entries;i++)
+    count[i] = 0;
+
+  for (i=0;i<nb_vectors;i++)
+  {
+    float dist;
+    nearest[i] = find_nearest_multi(codebook, nb_entries, data+MULTI*i*ndim, ndim, &dist);
+    err += dist;
+  }
+  printf("RMS error = %f\n", sqrt(err/nb_vectors/ndim));
+  for (i=0;i<nb_entries*ndim;i++)
+    codebook[i] = 0;
+
+  for (i=0;i<nb_vectors;i++)
+  {
+    int n = nearest[i];
+    count[n]++;
+    for (j=0;j<ndim;j++)
+      codebook[n*ndim+j] += data[(MULTI*i + (n&MULTI_MASK))*ndim+j];
+  }
+
+  float w2=0;
+  for (i=0;i<nb_entries;i++)
+  {
+    for (j=0;j<ndim;j++)
+      codebook[i*ndim+j] *= (1./count[i]);
+    w2 += (count[i]/(float)nb_vectors)*(count[i]/(float)nb_vectors);
+  }
+  //fprintf(stderr, "%f / %d\n", 1./w2, nb_entries);
+}
+
+
 void update_weighted(float *data, float *weight, int nb_vectors, float *codebook, int nb_entries, int ndim)
 {
   int i,j;
@@ -271,6 +338,38 @@
     update(data, nb_vectors, codebook, e, ndim);
 }
 
+void vq_train_multi(float *data, int nb_vectors, float *codebook, int nb_entries, int ndim)
+{
+  int i, j, e;
+  for (e=0;e<MULTI;e++) {
+    for (j=0;j<ndim;j++)
+      codebook[e*ndim+j] = 0;
+    for (i=0;i<nb_vectors;i++)
+      for (j=0;j<ndim;j++)
+        codebook[e*ndim+j] += data[(MULTI*i+e)*ndim+j];
+    for (j=0;j<ndim;j++) {
+      float delta = .01*(rand()/(float)RAND_MAX-.5);
+      codebook[e*ndim+j] *= (1./nb_vectors);
+      codebook[e*ndim+j] += delta;
+    }
+  }
+  e = MULTI;
+  for (j=0;j<10;j++)
+    update_multi(data, nb_vectors, codebook, e, ndim);
+
+  while (e < nb_entries)
+  {
+    split(codebook, e, ndim);
+    e<<=1;
+    fprintf(stderr, "%d\n", e);
+    for (j=0;j<4;j++)
+      update_multi(data, nb_vectors, codebook, e, ndim);
+  }
+  for (j=0;j<ndim*2;j++)
+    update_multi(data, nb_vectors, codebook, e, ndim);
+}
+
+
 void vq_train_weighted(float *data, float *weight, int nb_vectors, float *codebook, int nb_entries, int ndim)
 {
   int i, j, e;
@@ -303,8 +402,9 @@
 int main(int argc, char **argv)
 {
   int i,j;
-  int nb_vectors, nb_entries, ndim, ndim0, total_dim;
-  float *data, *pred, *codebook, *codebook2;
+  int nb_vectors, nb_entries, nb_entries2, ndim, ndim0, total_dim;
+  float *data, *pred, *multi_data, *multi_data2, *qdata;
+  float *codebook, *codebook2, *codebook_diff2, *codebook_diff4;
   float *delta;
   double err;
   FILE *fout;
@@ -314,11 +414,17 @@
   total_dim = atoi(argv[2]);
   nb_vectors = atoi(argv[3]);
   nb_entries = 1<<atoi(argv[4]);
+  nb_entries2 = 64;
   
   data = malloc((nb_vectors*ndim+total_dim)*sizeof(*data));
+  qdata = malloc((nb_vectors*ndim+total_dim)*sizeof(*qdata));
   pred = malloc(nb_vectors*ndim0*sizeof(*pred));
+  multi_data = malloc(MULTI*nb_vectors*ndim*sizeof(*multi_data));
+  multi_data2 = malloc(MULTI*nb_vectors*ndim*sizeof(*multi_data));
   codebook = malloc(nb_entries*ndim0*sizeof(*codebook));
   codebook2 = malloc(nb_entries*ndim0*sizeof(*codebook2));
+  codebook_diff4 = malloc(nb_entries*ndim*sizeof(*codebook_diff4));
+  codebook_diff2 = malloc(nb_entries2*ndim*sizeof(*codebook_diff2));
   
   for (i=0;i<nb_vectors;i++)
   {
@@ -348,8 +454,10 @@
   for (i=0;i<nb_vectors;i++)
   {
     int nearest = find_nearest(codebook, nb_entries, &pred[i*ndim0], ndim0, NULL);
+    qdata[i*ndim+j] = data[i*ndim+j];
     for (j=0;j<ndim0;j++)
     {
+      qdata[i*ndim+j+1] = codebook[nearest*ndim0+j];
       delta[i*ndim0+j] = pred[i*ndim0+j] - codebook[nearest*ndim0+j];
       err += delta[i*ndim0+j]*delta[i*ndim0+j];
     }
@@ -366,12 +474,44 @@
     n1 = find_nearest(codebook2, nb_entries, &delta[i*ndim0], ndim0, NULL);
     for (j=0;j<ndim0;j++)
     {
-      delta[i*ndim0+j] = delta[i*ndim0+j] - codebook2[n1*ndim0+j];
+      qdata[i*ndim+j+1] += codebook2[n1*ndim0+j];
+      //delta[i*ndim0+j] = delta[i*ndim0+j] - codebook2[n1*ndim0+j];
+      delta[i*ndim0+j] = qdata[i*ndim+j+1] - data[i*ndim+j+1];
       err += delta[i*ndim0+j]*delta[i*ndim0+j];
     }
   }
   fprintf(stderr, "Cepstrum RMS error after stage 2: %f)\n", sqrt(err/nb_vectors/ndim));
-  
+
+  for (i=0;i<nb_vectors-4;i++)
+  {
+    for (j=0;j<ndim0;j++)
+      multi_data[MULTI*i*ndim+j] = data[(i+1)*ndim+j+1] - .5*(qdata[i*ndim+j+1]+qdata[(i+2)*ndim+j+1]);
+    for (j=0;j<ndim0;j++)
+      multi_data[(MULTI*i+1)*ndim+j] = data[(i+1)*ndim+j+1] - .5*(qdata[i*ndim+j+1]+qdata[(i+2)*ndim+j+1]);
+    for (j=0;j<ndim0;j++)
+      multi_data[(MULTI*i+2)*ndim+j] = data[(i+1)*ndim+j+1] - qdata[i*ndim+j+1];
+    for (j=0;j<ndim0;j++)
+      multi_data[(MULTI*i+3)*ndim+j] = data[(i+1)*ndim+j+1] - qdata[(i+2)*ndim+j+1];
+  }
+
+  for (i=0;i<nb_vectors-4;i++)
+  {
+    for (j=0;j<ndim0;j++)
+      multi_data2[MULTI*i*ndim+j] = data[(i+2)*ndim+j+1] - .5*(qdata[i*ndim+j+1]+qdata[(i+4)*ndim+j+1]);
+    for (j=0;j<ndim0;j++)
+      multi_data2[(MULTI*i+1)*ndim+j] = data[(i+2)*ndim+j+1] - .5*(qdata[i*ndim+j+1]+qdata[(i+4)*ndim+j+1]);
+    for (j=0;j<ndim0;j++)
+      multi_data2[(MULTI*i+2)*ndim+j] = data[(i+2)*ndim+j+1] - qdata[i*ndim+j+1];
+    for (j=0;j<ndim0;j++)
+      multi_data2[(MULTI*i+3)*ndim+j] = data[(i+2)*ndim+j+1] - qdata[(i+4)*ndim+j+1];
+  }
+
+  vq_train_multi(multi_data2, nb_vectors-4, codebook_diff4, nb_entries, ndim);
+
+  printf("done\n");
+  vq_train_multi(multi_data, nb_vectors-4, codebook_diff2, 64, ndim);
+
+
   fout = fopen("ceps_codebooks.c", "w");
   fprintf(fout, "/* This file is automatically generated */\n\n");
   fprintf(fout, "float ceps_codebook1[%d*%d] = {\n",nb_entries, ndim0);
@@ -385,11 +525,28 @@
   fprintf(fout, "};\n\n");
 
   fprintf(fout, "float ceps_codebook2[%d*%d] = {\n",nb_entries, ndim0);
-  
   for (i=0;i<nb_entries;i++)
   {
     for (j=0;j<ndim0;j++)
       fprintf(fout, "%f, ", codebook2[i*ndim0+j]);
+    fprintf(fout, "\n");
+  }
+  fprintf(fout, "};\n\n");
+
+  fprintf(fout, "float ceps_codebook_diff4[%d*%d] = {\n",nb_entries, ndim);
+  for (i=0;i<nb_entries;i++)
+  {
+    for (j=0;j<ndim;j++)
+      fprintf(fout, "%f, ", codebook_diff4[i*ndim+j]);
+    fprintf(fout, "\n");
+  }
+  fprintf(fout, "};\n\n");
+
+  fprintf(fout, "float ceps_codebook_diff2[%d*%d] = {\n",nb_entries2, ndim);
+  for (i=0;i<nb_entries2;i++)
+  {
+    for (j=0;j<ndim;j++)
+      fprintf(fout, "%f, ", codebook_diff2[i*ndim+j]);
     fprintf(fout, "\n");
   }
   fprintf(fout, "};\n\n");
-- 
⑨