feat(sv56): variable bit-width support (8-32 bit) for speech voltmeter

.gitignore

@@ -37,6 +37,8 @@ sweep.[aru]*
 *.tst
 !*voic*.tst
 
+src/**/test_data/
+
 src/g722/test_data/inpsp.bs
 src/g722/test_data/outsp.e-d
 src/g727/test_data/rn54_a.out
@@ -97,3 +99,8 @@ bld/
 .vs/
 *vcxproj*
 *.sln*
+.vscode/
+/build
+*.double
+stderr.txt
+stdout.txt

CMakeLists.txt

@@ -1,6 +1,8 @@
-cmake_minimum_required(VERSION 3.1)
+cmake_minimum_required(VERSION 3.10)
+
 project(STL2023 VERSION 2023)
-add_definitions( -DVERSION_STL="${CMAKE_PROJECT_NAME}" )
+
+add_definitions( -DVERSION_STL="${CMAKE_PROJECT_NAME}" -DCMAKE_POLICY_VERSION_MINIMUM=3.5 )
 
 set(CMAKE_RUNTIME_OUTPUT_DIRECTORY ${CMAKE_BINARY_DIR}/bin)
 

doc/manual/sv56.tex

@@ -224,9 +224,11 @@ \section{Implementation}
 {\em n}.}:
 \begin{quote} \normalsize
  {\em f}        \hfill \parbox{100mm}{\SF Sampling frequency, in Hz }\\
- {\em a[15]}    \hfill \parbox{100mm}{\SF Activity count }\\
- {\em c[15]}    \hfill \parbox{100mm}{\SF Threshold level }\\
- {\em hang[15]} \hfill \parbox{100mm}{\SF Hangover count }\\
+ {\em a[SVP56\_MAX\_THRESHOLDS]}    \hfill \parbox{100mm}{\SF Activity count }\\
+ {\em c[SVP56\_MAX\_THRESHOLDS]}    \hfill \parbox{100mm}{\SF Threshold level }\\
+ {\em hang[SVP56\_MAX\_THRESHOLDS]} \hfill \parbox{100mm}{\SF Hangover count }\\
+ {\em thres\_no} \hfill \parbox{100mm}{\SF Actual number of thresholds in use (bitno$-$1) }\\
+ {\em bitno}     \hfill \parbox{100mm}{\SF Bit depth of input signal (default 16) }\\
  {\em n}        \hfill \parbox{100mm}{\SF Number of samples read since last  
                                           reset }\\
  {\em s}        \hfill \parbox{100mm}{\SF Sum of all samples since last   
@@ -270,7 +272,7 @@ \subsection{{\tt init\_speech\_voltmeter}}
 \#include "sv-p56.h"\\
 void init\_speech\_voltmeter
          \pbox{110mm}{
-            (SVP56\_state {\em *state}, double {\em f});
+            (SVP56\_state {\em *state}, double {\em f}, int {\em bitno});
          }
 }
 
@@ -280,13 +282,20 @@ \subsection{{\tt init\_speech\_voltmeter}}
 
 {\tt init\_speech\_voltmeter} performs the initialization of the
 speech voltmeter state variables in the structure pointed by {\em
-state} to the appropriate initial values. The only value required from
-the user is the sampling rate $f$ (in Hz) of the signal that the speech
-voltmeter is supposed to measure. Note that when measuring new
+state} to the appropriate initial values. The values required from
+the user are the sampling rate $f$ (in Hz) of the signal that the speech
+voltmeter is supposed to measure, and the bit depth {\em bitno} of the
+input signal. Note that when measuring new
 speech material, the state variable shall be re-initialized, otherwise
 accumulation of previous measurements will happen and wrong measurements 
 will be reported.
 
+The header file {\tt sv-p56.h} defines two constants that control the
+maximum bit depth supported:
+\begin{quote} \normalsize
+ {\tt SVP56\_MAX\_NO\_BITS} \hfill \parbox{100mm}{\SF Maximum bit depth supported (default: 32). }\\
+ {\tt SVP56\_MAX\_THRESHOLDS} \hfill \parbox{100mm}{\SF Maximum number of thresholds ($=$ SVP56\_MAX\_NO\_BITS $-$ 1). }
+\end{quote}
 
 
 {\bf Variables: }
@@ -298,6 +307,14 @@ \subsection{{\tt init\_speech\_voltmeter}}
                Is the sampling rate (in Hz) of the signal to be measured in
                the next calls of {\tt speech\_voltmeter}. If zero or 
                negative, the sampling rate is initialized to 16000 Hz.
+
+\item[\pbox{20mm}{\em bitno}] %\rulex{1mm}\\
+               Is the bit depth (resolution) of the input signal. Typical
+               values are 16 (default for narrowband speech) and 24 or 32
+               for higher-resolution signals. This parameter sets the number
+               of thresholds used by the algorithm to {\em bitno $-$ 1}, and
+               initializes the peak detectors {\em maxP} and {\em maxN} to
+               $\pm 2^{bitno-1}$. Must not exceed {\tt SVP56\_MAX\_NO\_BITS}.
 \end{Descr}
 
 {\bf Return value: }      
@@ -435,25 +452,31 @@ \section{Portability and compliance}
 input file, which is saved in an aoutput file. Levels are reported in
 dBov.
 
-In general, input files are in integer representation, 16-bit words,
-2's complement (i.e., {\tt short} data). In UGST convention, this
-data must be left-adjusted, {\em rather} than right-adjusted. Since
-the speech voltmeter uses {\tt float} input data, it is necessary to
-convert from {\tt short} (in the mentioned format) to {\tt float};
-this is carried out by the function {\tt sh2fl()}. In addition, the
-option to `normalize' the input data to the range -1..+1 is selected.
+The example programs support variable bit-width input files from 8 to
+32 bits per sample (controlled by the {\em bitno} parameter). Supported
+storage widths are 8-bit (1 byte/sample), 16-bit (2 bytes/sample),
+24-bit (3 bytes/sample), and 32-bit (4 bytes/sample). For bit depths
+between 12 and 15, samples are stored in 2 bytes with the least
+significant bits masked to zero. Files use native byte order.
+
+Since the speech voltmeter uses {\tt float} input data normalized to
+the range --1.0..+1.0, the helper functions in {\tt sv56-util.h} handle
+the conversion between raw integer samples and normalized floats:
+{\tt sv56\_raw2fl()} reads raw bytes and normalizes by dividing by
+$2^{bitno-1}$, and {\tt sv56\_fl2raw()} performs the inverse operation
+with truncation and hard-clipping.
+
 After the equalization factor is found, results are reported on the
 screen, which varies according to the program used and some of the
-command-line options. 
+command-line options.
 
 While program {\tt actlevel.c} stops at this point, program {\tt
 sv56demo.c} proceeds calling the function {\tt scale()} to carry out
 the (amplitude) equalization using single (rather than double) float
 precision. After equalization, the samples are converted back to
-integer (short, right-justified) with the routine {\tt fl2sh()} using
-truncation, no zero-padding of the least significant bits,
-left-justification of data, and hard-clipping of data above the
-overload point. After that, data is saved to the user-specified file .
+raw integer format with {\tt sv56\_fl2raw()} using
+truncation and hard-clipping of data above the
+overload point. After that, data is saved to the user-specified file.
 
 
 %-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-.-
@@ -471,7 +494,7 @@ \section{Portability and compliance}
 #include <math.h>
 #include "ugstdemo.h"           /* ... UGST demonstration program defs ... */
 #include "sv-p56.h"             /* ... SV-P56 prototypes & defs ... */
-#include "ugst-utl.h"           /* ... UGST utilities ... */
+#include "sv56-util.h"          /* ... SV56 I/O utilities ... */
 #define BLK_LEN 256
 
 void main(argc, argv)
@@ -482,7 +505,8 @@ \section{Portability and compliance}
   char            FileIn[180];  /* input file name */
   FILE           *Fi;           /* input file pointers */
   long            N=BLK_LEN, l;
-  short           bitno, buffer[BLK_LEN];
+  int             bitno, bps;
+  unsigned char   raw_buf[BLK_LEN * 4];
   float           Buf[BLK_LEN];
   double          ActiveLeveldB, sf, satur;
 
@@ -493,20 +517,21 @@ \section{Portability and compliance}
   FIND_PAR_D(2, "_Sampling Frequency: .. ", sf, 16000);
   FIND_PAR_L(3, "_A/D resolution: ...... ", bitno, 16);
 
-  /* Calculate overload point in the non-normalized range */
+  /* Calculate bytes per sample and overload point */
+  bps = sv56_bytes_per_sample(bitno);
   satur = pow ((double)2.0, (double)(bitno - 1));
 
   /* Reset- variables for speech level measurements */
-  init_speech_voltmeter(&state, sf);
+  init_speech_voltmeter(&state, sf, bitno);
 
   /* Opening input file */
   Fi = fopen(FileIn, RB);
 
   /* Read samples ... */
-  while ((l = fread(buffer, N, sizeof(short), Fi)) > 0)
+  while ((l = fread(raw_buf, bps, N, Fi)) > 0)
   {
-    /* ... Convert samples to float, normalizing to +1..-1 */
-    sh2fl((long) l, buffer, Buf, (long) state.bitno, 1);
+    /* ... Convert raw samples to float, normalizing to +1..-1 */
+    sv56_raw2fl((long) l, raw_buf, Buf, bitno);
 
     /* ... Get the active level */
     ActiveLeveldB = speech_voltmeter(Buf, (long) l, &state);
@@ -524,7 +549,7 @@ \section{Portability and compliance}
     printf("\n  Maximum negative value: .. %7.0f [PCM]", 
            SVP56_get_neg_max(state) * satur);
     printf("\n  Long-term energy (rms): .. %7.3f [dBov]", 
-           SVP56_get_rms_dB(state);
+           SVP56_get_rms_dB(state));
     printf("\n  Active speech level: ..... %7.3f [dBov]", ActiveLeveldB);
     printf("\n  Activity factor: ......... %7.3f [%%]",
            SVP56_get_activity(state));

src/sv56/CMakeLists.txt

@@ -8,10 +8,40 @@ add_executable(actlev actlevel.c  sv-p56.c ../utl/ugst-utl.c)
 target_link_libraries(actlev ${M_LIBRARY})
 
 add_test(sv56demo1 ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/sv56demo -q test_data/voice.src test_data/voice.prc 256 1 0 -30)
-add_test(sv56demo1-verify ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/cf -q test_data/voice.nrm test_data/voice.prc)
+set_tests_properties(sv56demo1 PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56demo1-verify ${CMAKE_COMMAND} -E compare_files test_data/voice.nrm test_data/voice.prc)
+set_tests_properties(sv56demo1-verify PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
 
 add_test(sv56demo2 ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/sv56demo -q -rms test_data/voice.src test_data/voice.rms 256 1 0 -30)
-add_test(sv56demo2-verify ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/cf -q test_data/voice.ltl test_data/voice.rms)
+set_tests_properties(sv56demo2 PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56demo2-verify ${CMAKE_COMMAND} -E compare_files test_data/voice.ltl test_data/voice.rms)
+set_tests_properties(sv56demo2-verify PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
 
 add_test(sv56demo3 ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/actlev -q test_data/voice.src test_data/voice.nrm test_data/voice.prc test_data/voice.ltl test_data/voice.rms)
+set_tests_properties(sv56demo3 PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_executable(sv56test sv56test.c sv-p56.c)
+target_link_libraries(sv56test ${M_LIBRARY})
+
+add_executable(sv56-convert sv56-convert.c)
+target_link_libraries(sv56-convert ${M_LIBRARY})
+
+add_test(sv56test-16bit ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/sv56test 16 test_data/voice.src test_data/voice_16bit.out)
+set_tests_properties(sv56test-16bit PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56test-16bit-verify ${CMAKE_COMMAND} -E compare_files test_data/voice_16bit.ref test_data/voice_16bit.out)
+set_tests_properties(sv56test-16bit-verify PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56test-24bit ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/sv56test 24 test_data/voice_24.src test_data/voice_24bit.out)
+set_tests_properties(sv56test-24bit PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56test-24bit-verify ${CMAKE_COMMAND} -E compare_files test_data/voice_24bit.ref test_data/voice_24bit.out)
+set_tests_properties(sv56test-24bit-verify PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
+
+add_test(sv56test-32bit ${CMAKE_RUNTIME_OUTPUT_DIRECTORY}/sv56test 32 test_data/voice_32.src test_data/voice_32bit.out)
+set_tests_properties(sv56test-32bit PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})
 
+add_test(sv56test-32bit-verify ${CMAKE_COMMAND} -E compare_files test_data/voice_32bit.ref test_data/voice_32bit.out)
+set_tests_properties(sv56test-32bit-verify PROPERTIES WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR})

src/sv56/actlevel.c

@@ -155,6 +155,7 @@
 
 /* ... Include of utilities ... */
 #include "ugst-utl.h"
+#include "sv56-util.h"
 
 /* ... Local definitions ... */
 #define DEF_BLK_LEN 256         /* samples per block */
@@ -430,8 +431,9 @@ int main (int argc, char *argv[]) {
 #endif
 
   /* Other variables */
-  short buffer[4096];
+  unsigned char raw_buf[4096 * 4];
   float Buf[4096];
+  int bps;
   long start_byte, bitno = 16;
   double sf = 16000;            /* Hz */
   double ActiveLeveldB, level = 0, gain = 0;
@@ -538,9 +540,16 @@ int main (int argc, char *argv[]) {
 
 
   /* ......... SOME INITIALIZATIONS ......... */
+  /* Validate bitno */
+  if (bitno < 8 || bitno > SVP56_MAX_NO_BITS) {
+    fprintf (stderr, "Error: bitno must be between 8 and %d\n", SVP56_MAX_NO_BITS);
+    exit (1);
+  }
+  bps = sv56_bytes_per_sample ((int) bitno);
+
   /* funny_size = strlen(funny); */
   start_byte = --N1;
-  start_byte *= N * sizeof (short);
+  start_byte *= N * bps;
   N2_ori = N2;
 
   /* Overflow (saturation) point */
@@ -555,13 +564,13 @@ int main (int argc, char *argv[]) {
     argc--;
 
     /* Reset variables for speech level measurements */
-    init_speech_voltmeter (&state, sf);
+    init_speech_voltmeter (&state, sf, (int)bitno);
 
     /* ......... FILE PREPARATION ......... */
 
     /* Opening input file; abort if there's any problem */
 #ifdef VMS
-    sprintf (mrs, "mrs=%d", 2 * N);
+    sprintf (mrs, "mrs=%d", bps * N);
 #endif
     if ((Fi = fopen (FileIn, RB)) == NULL)
       KILL (FileIn, 2);
@@ -573,7 +582,7 @@ int main (int argc, char *argv[]) {
     if (N2 == 0) {
       struct stat st;
       stat (FileIn, &st);
-      N2 = ceil (st.st_size / (double) (N * sizeof (short)));
+      N2 = ceil (st.st_size / (double) (N * bps));
     }
 
     /* Move pointer to 1st block of interest */
@@ -587,9 +596,9 @@ int main (int argc, char *argv[]) {
     if (!quiet)
       fprintf (stderr, "  Processing \r");
     for (i = 0; i < N2; i++) {
-      if ((l = fread (buffer, sizeof (short), N, Fi)) > 0) {
+      if ((l = fread (raw_buf, bps, N, Fi)) > 0) {
         /* ... Convert samples to float */
-        sh2fl ((long) l, buffer, Buf, bitno, 1);
+        sv56_raw2fl ((long) l, raw_buf, Buf, (int) bitno);
 
         /* ... Get the active level */
         ActiveLeveldB = speech_voltmeter (Buf, (long) l, &state);

src/sv56/sv-p56.c

@@ -223,21 +223,23 @@ double bin_interp (double upcount, double lwcount, double upthr, double lwthr, d
 #define M        15.9           /* in [dB] */
 #define THRES_NO 15             /* number of thresholds in the speech voltmeter */
 
-void init_speech_voltmeter (SVP56_state * state, double sampl_freq) {
+void init_speech_voltmeter (SVP56_state * state, double sampl_freq, int bitno) {
   double x;
   long I, j;
 
 
   /* First initializations */
   state->f = sampl_freq;
+  state->bitno = bitno;
+  state->thres_no = bitno - 1;
   I = floor (H * state->f + 0.5);
 
-  /* Inicialization of threshold vector */
-  for (x = 0.5, j = 1; j <= THRES_NO; j++, x /= 2.0)
-    state->c[THRES_NO - j] = x;
+  /* Inicialization of threshold vector: geometric progression from 0.5 down to 2^-(bitno-1) */
+  for (x = 0.5, j = 1; j <= state->thres_no; j++, x /= 2.0)
+    state->c[state->thres_no - j] = x;
 
   /* Inicialization of activity and hangover count vectors */
-  for (j = 0; j < THRES_NO; j++) {
+  for (j = 0; j < state->thres_no; j++) {
     state->a[j] = 0;
     state->hang[j] = I;
   }
@@ -247,8 +249,8 @@ void init_speech_voltmeter (SVP56_state * state, double sampl_freq) {
 
   /* Inicialization of other quantities referring to state variables */
   state->max = 0;
-  state->maxP = -32768.;
-  state->maxN = 32767.;
+  state->maxP = -pow(2.0, (double)(bitno - 1));
+  state->maxN = pow(2.0, (double)(bitno - 1)) - 1.0;
 
   /* Defining the 0 dB reference level in terms of normalized values */
   state->refdB = 0 /* dBov */ ;
@@ -350,7 +352,7 @@ double speech_voltmeter (float *buffer, long smpno, SVP56_state * state) {
   int I, j;
   long k;
   double g, x, AdB, CdB, AmdB, CmdB, ActiveSpeechLevel;
-  double LongTermLevel, Delta[15];
+  double LongTermLevel, Delta[SVP56_MAX_THRESHOLDS];
 
 
   /* Some initializations */
@@ -380,7 +382,7 @@ double speech_voltmeter (float *buffer, long smpno, SVP56_state * state) {
     state->q = g * (state->q) + (1 - g) * (state->p);
 
     /* Applies threshold to the envelope q */
-    for (j = 0; j < THRES_NO; j++) {
+    for (j = 0; j < state->thres_no; j++) {
       if ((state->q) >= state->c[j]) {
         state->a[j]++;
         state->hang[j] = 0;
@@ -412,7 +414,7 @@ double speech_voltmeter (float *buffer, long smpno, SVP56_state * state) {
     return (ActiveSpeechLevel);
 
   /* Proceed serially for steps 2 and up -- this is the most common case */
-  for (j = 1; j < THRES_NO; j++) {
+  for (j = 1; j < state->thres_no; j++) {
     if (state->a[j] != 0) {
       AdB = 10 * log10 (((state->sq) / state->a[j]) + MIN_LOG_OFFSET);
       CdB = 20 * log10 (((double) state->c[j]) + MIN_LOG_OFFSET);