Audio pre-roll and loudness measurement support for USAC (#79)

Tests done: Smoke test

Android.bp

@@ -430,6 +430,7 @@ cc_library_static {
         "encoder/ixheaace_hybrid.c",
         "encoder/ixheaace_hybrid_init.c",
         "encoder/ixheaace_interface.c",
+        "encoder/ixheaace_loudness_measurement.c",
         "encoder/ixheaace_mdct_480.c",
         "encoder/ixheaace_mps_bitstream.c",
         "encoder/ixheaace_mps_dct.c",

README_enc.md

@@ -52,6 +52,8 @@ The configuration file for DRC is placed in `encoder\test` directory(impd_drc_co
 [-esbr_hq:<esbr_hq_flag>]
 [-drc:<drc_flag>]
 [-inter_tes_enc:<inter_tes_enc_flag>]
+[-rap:<random access interval in ms>]
+[-stream_id:<stream identifier>]
 
 where,
   <paramfile> is the parameter file with multiple commands
@@ -101,6 +103,9 @@ where,
   <esbr_hq_flag> Valid values are 0 (disable high quality eSBR) and 1 (enable high quality eSBR). Default is 0.
   <drc_flag> Valid values are 0 (disable DRC encoding) and 1 (enable DRC encoding). Default is 0.
   <inter_tes_enc_flag> Valid values are 0 (disable inter-TES encoding) and 1 (enable inter-TES encoding). Default is 0.
+  <random access interval in ms> is the time interval between audio preroll frames in ms. It is applicable only for AOT 42. Valid values are -1 (Audio preroll sent only at beginning of file) and greater than 1000 ms. Default is -1.
+  <stream identifier> It is the stream id used to uniquely identify configuration of a stream within a set of associated streams. It is applicable only for AOT 42. Valid values are 0 to 65535. Any value outside this range is type-casted to a value of unsigned short type. Default is 0.
+
 ```
 Sample CLI:
 ```

encoder/drc_src/impd_drc_api.c

@@ -18,6 +18,7 @@
  * Originally developed and contributed by Ittiam Systems Pvt. Ltd, Bangalore
  */
 
+#include <string.h>
 #include "ixheaac_type_def.h"
 #include "ixheaac_error_standards.h"
 #include "ixheaace_error_codes.h"
@@ -311,6 +312,22 @@ IA_ERRORCODE impd_drc_enc_init(VOID *pstr_drc_state, VOID *ptr_drc_scratch,
   return err_code;
 }
 
+IA_ERRORCODE impd_loudness_info_init(VOID *pstr_drc_state, ia_drc_input_config *pstr_inp_config) {
+  IA_ERRORCODE err_code = IA_NO_ERROR;
+  ia_drc_enc_state *pstr_drc_state_local = pstr_drc_state;
+
+  iusace_create_bit_buffer(&pstr_drc_state_local->str_bit_buf_cfg_ext,
+    pstr_drc_state_local->bit_buf_base_cfg_ext,
+    sizeof(pstr_drc_state_local->bit_buf_base_cfg_ext), 1);
+
+  memcpy(&pstr_drc_state_local->str_gain_enc.str_loudness_info_set,
+    &pstr_inp_config->str_enc_loudness_info_set,
+    sizeof(ia_drc_loudness_info_set_struct));
+
+  err_code = impd_drc_write_measured_loudness_info(pstr_drc_state_local);
+  return err_code;
+}
+
 IA_ERRORCODE impd_drc_enc(VOID *pstr_drc_state, FLOAT32 **pptr_input, UWORD32 inp_offset,
                           WORD32 *ptr_bits_written, VOID *pstr_scratch) {
   LOOPIDX i, j, k;

encoder/drc_src/impd_drc_api.h

@@ -36,8 +36,17 @@ typedef struct {
   ia_drc_uni_drc_gain_ext_struct str_enc_gain_extension;
 } ia_drc_input_config;
 
+typedef struct {
+  ia_drc_enc_params_struct str_enc_params;
+  ia_drc_uni_drc_config_struct str_uni_drc_config;
+  ia_drc_loudness_info_set_struct str_enc_loudness_info_set;
+  ia_drc_uni_drc_gain_ext_struct str_enc_gain_extension;
+} ia_drc_internal_config;
+
 IA_ERRORCODE impd_drc_enc_init(VOID *pstr_drc_state, VOID *ptr_drc_scratch,
                                ia_drc_input_config *pstr_inp_config);
 
 IA_ERRORCODE impd_drc_enc(VOID *pstr_drc_state, FLOAT32 **pptr_input, UWORD32 inp_offset,
                           WORD32 *ptr_bits_written, VOID *pstr_scratch);
+
+IA_ERRORCODE impd_loudness_info_init(VOID *pstr_drc_state, ia_drc_input_config *pstr_inp_config);

encoder/drc_src/impd_drc_enc.h

@@ -38,3 +38,9 @@ IA_ERRORCODE impd_drc_write_loudness_info_set_extension(
 IA_ERRORCODE impd_drc_write_uni_drc_config(ia_drc_enc_state *pstr_drc_state, WORD32 *ptr_bit_cnt);
 
 VOID impd_drc_write_uni_drc_gain(ia_drc_enc_state *pstr_drc_state, WORD32 *ptr_bit_cnt);
+
+IA_ERRORCODE impd_drc_write_measured_loudness_info(ia_drc_enc_state *pstr_drc_state);
+
+IA_ERRORCODE impd_drc_write_loudness_info_set(ia_drc_enc_state *pstr_drc_state,
+  ia_bit_buf_struct *it_bit_buf,
+  WORD32 *ptr_bit_cnt);

encoder/drc_src/impd_drc_mux.c

@@ -3005,6 +3005,20 @@ IA_ERRORCODE impd_drc_write_uni_drc_config(ia_drc_enc_state *pstr_drc_state,
   return err_code;
 }
 
+IA_ERRORCODE impd_drc_write_measured_loudness_info(ia_drc_enc_state *pstr_drc_state) {
+
+  IA_ERRORCODE err_code = IA_NO_ERROR;
+  ia_bit_buf_struct *it_bit_buf_lis = &pstr_drc_state->str_bit_buf_cfg_ext;
+  WORD32 bit_cnt_lis = 0;
+  err_code = impd_drc_write_loudness_info_set(pstr_drc_state, it_bit_buf_lis, &bit_cnt_lis);
+  if (err_code & IA_FATAL_ERROR) {
+    return (err_code);
+  }
+  pstr_drc_state->drc_config_ext_data_size_bit = bit_cnt_lis;
+
+  return err_code;
+}
+
 IA_ERRORCODE impd_drc_enc_initial_gain(const WORD32 gain_coding_profile, FLOAT32 gain_initial,
                                        FLOAT32 *gain_initial_quant, WORD32 *code_size,
                                        WORD32 *code) {

encoder/drc_src/impd_drc_uni_drc.h

@@ -130,6 +130,8 @@
 #define MIN_METHOD_VALUE (-116.0f)
 #define MAX_METHOD_VALUE (121.0f)
 #define MAX_FLT_VAL_DB (770.6367883810890080451095799195f)
+#define DEFAULT_METHOD_VALUE (-31.0f)
+#define DEFAULT_SAMPLE_PEAK_VALUE (-31.0f)
 
 typedef struct {
   WORD32 level_estim_k_weighting_type;

encoder/iusace_cnst.h

@@ -201,3 +201,12 @@
 #define USACE_MAX_SCR_SIZE (733836)
 #define USACE_SCR_STACK (10 * 1024)
 #define MAX_USAC_ESBR_BITRATE (96000)
+
+#define MAX_PREROLL_FRAMES (3)
+#define MAX_OUTPUT_BYTES_PER_CH (768)
+#define MAXIMUM_VALUE_8BIT (255)
+#define DEFAULT_RAP_INTERVAL_IN_MS (-1)
+#define MIN_RAP_INTERVAL_IN_MS (1000)
+#define MAX_PREROLL_CONFIG_SIZE (1024)
+#define CC_NUM_PREROLL_FRAMES (1)
+#define USAC_FIRST_FRAME_FLAG_DEFAULT_VALUE (1)

encoder/iusace_config.h

@@ -143,10 +143,13 @@ typedef struct {
 
 #define ID_EXT_ELE_FILL 0
 #define ID_EXT_ELE_UNI_DRC 4
+#define ID_EXT_ELE_AUDIOPREROLL (3)
 
 #define ID_CONFIG_EXT_FILL 0
 #define ID_CONFIG_EXT_DOWNMIX (1)
 #define ID_CONFIG_EXT_LOUDNESS_INFO (2)
+#define ID_CONFIG_EXT_STREAM_ID (7)
+#define CONFIG_EXT_LEN_STREAM_ID (2)
 #define NUM_COEFF (1024)
 
 typedef enum {
@@ -223,6 +226,7 @@ typedef struct {
   WORD32 output_channel_pos[BS_MAX_NUM_OUT_CHANNELS];
   WORD32 ccfl;
   ia_usac_enc_element_config_struct str_usac_element_config[USAC_MAX_ELEMENTS];
+  UWORD16 stream_identifier;
 } ia_usac_config_struct;
 
 typedef struct {
@@ -272,6 +276,16 @@ typedef struct {
   WORD32 drc_frame_size;
   ia_drc_input_config str_drc_cfg;
   WORD32 use_acelp_only;
+  WORD32 random_access_interval;
+  WORD32 preroll_flag;
+  WORD32 num_preroll_frames;
+  WORD32 preroll_idx;
+  WORD32 is_ipf;
+  WORD32 preroll_frame;
+  WORD32 is_first_frame;
+  ia_drc_internal_config str_internal_drc_cfg;
+  WORD32 use_measured_loudness;
+  UWORD16 stream_id;
 } ia_usac_encoder_config_struct;
 
 typedef struct {

encoder/iusace_enc_main.c

@@ -618,7 +618,14 @@ IA_ERRORCODE iusace_enc_init(ia_usac_encoder_config_struct *ptr_usac_config,
       p_audio_specific_config->channel_configuration = ptr_usac_config->channels;
     }
   }
-
+  ia_usac_enc_element_config_struct *pstr_usac_elem_config =
+      &(pstr_asc_usac_config->str_usac_element_config[pstr_asc_usac_config->num_elements]);
+  pstr_asc_usac_config->usac_element_type[pstr_asc_usac_config->num_elements] = ID_USAC_EXT;
+  pstr_usac_elem_config->usac_ext_ele_type = ID_EXT_ELE_AUDIOPREROLL;
+  pstr_usac_elem_config->usac_ext_ele_dflt_len_present = 0;
+  pstr_usac_elem_config->usac_ext_ele_payload_present = 0;
+  pstr_usac_elem_config->usac_ext_ele_cfg_len = 0;
+  pstr_asc_usac_config->num_elements++;
   // DRC Config
   if (ptr_usac_config->use_drc_element) {
     ptr_usac_config->str_drc_cfg.str_uni_drc_config.str_channel_layout.base_ch_count =
@@ -637,26 +644,40 @@ IA_ERRORCODE iusace_enc_init(ia_usac_encoder_config_struct *ptr_usac_config,
 
     if (ptr_usac_config->use_drc_element) {
       ia_usac_enc_element_config_struct *pstr_usac_elem_config =
-        &(pstr_asc_usac_config->str_usac_element_config[pstr_asc_usac_config->num_elements]);
+          &(pstr_asc_usac_config->str_usac_element_config[pstr_asc_usac_config->num_elements]);
       pstr_asc_usac_config->usac_element_type[pstr_asc_usac_config->num_elements] = ID_USAC_EXT;
       pstr_usac_elem_config->usac_ext_ele_type = ID_EXT_ELE_UNI_DRC;
       pstr_usac_elem_config->usac_ext_ele_dflt_len_present = 0;
       pstr_usac_elem_config->usac_ext_ele_payload_present = 0;
       pstr_usac_elem_config->drc_config_data = usac_data->str_drc_state.bit_buf_base_cfg;
       pstr_usac_elem_config->usac_ext_ele_cfg_len =
-        (usac_data->str_drc_state.drc_config_data_size_bit + 7) >> 3;
+          (usac_data->str_drc_state.drc_config_data_size_bit + 7) >> 3;
       pstr_asc_usac_config->num_elements++;
     }
+  } else {
+    err_code =
+        impd_loudness_info_init(&usac_data->str_drc_state, &ptr_usac_config->str_drc_cfg);
+    if (err_code) {
+      return err_code;
+    }
   }
-  if (ptr_usac_config->use_drc_element)  // For Loudness
-  {
-    pstr_asc_usac_config->usac_config_ext_type[pstr_asc_usac_config->num_config_extensions] =
-        ID_CONFIG_EXT_LOUDNESS_INFO;
-    pstr_asc_usac_config->usac_config_ext_len[pstr_asc_usac_config->num_config_extensions] =
-        (usac_data->str_drc_state.drc_config_ext_data_size_bit + 7) >> 3;
-    pstr_asc_usac_config->usac_config_ext_buf[pstr_asc_usac_config->num_config_extensions] =
-        usac_data->str_drc_state.bit_buf_base_cfg_ext;
-    pstr_asc_usac_config->num_config_extensions++;
+
+  pstr_asc_usac_config->usac_config_ext_type[pstr_asc_usac_config->num_config_extensions] =
+      ID_CONFIG_EXT_STREAM_ID;
+  pstr_asc_usac_config->usac_config_ext_len[pstr_asc_usac_config->num_config_extensions] =
+      CONFIG_EXT_LEN_STREAM_ID;
+  pstr_asc_usac_config->num_config_extensions++;
+  pstr_asc_usac_config->stream_identifier = ptr_usac_config->stream_id;
+
+  pstr_asc_usac_config->usac_config_ext_type[pstr_asc_usac_config->num_config_extensions] =
+      ID_CONFIG_EXT_LOUDNESS_INFO;
+  pstr_asc_usac_config->usac_config_ext_len[pstr_asc_usac_config->num_config_extensions] =
+      (usac_data->str_drc_state.drc_config_ext_data_size_bit + 7) >> 3;
+  pstr_asc_usac_config->usac_config_ext_buf[pstr_asc_usac_config->num_config_extensions] =
+      usac_data->str_drc_state.bit_buf_base_cfg_ext;
+  pstr_asc_usac_config->num_config_extensions++;
+
+  if (pstr_asc_usac_config->num_config_extensions) {
     pstr_asc_usac_config->usac_cfg_ext_present = 1;
   }
 
@@ -778,7 +799,7 @@ IA_ERRORCODE iusace_enc_init(ia_usac_encoder_config_struct *ptr_usac_config,
 
   for (; elem_idx < pstr_asc_usac_config->num_elements; elem_idx++) {
     idx = elem_idx - pstr_asc_usac_config->num_ext_elements;
-    pstr_asc_usac_config->str_usac_element_config[idx].noise_filling =
+    pstr_asc_usac_config->str_usac_element_config[elem_idx].noise_filling =
         usac_data->noise_filling[idx];
     usac_data->channel_elem_type[idx] = pstr_asc_usac_config->usac_element_type[elem_idx];
   }
@@ -792,6 +813,8 @@ IA_ERRORCODE iusace_enc_init(ia_usac_encoder_config_struct *ptr_usac_config,
     pstr_usac_elem_config->usac_ext_ele_dflt_len_present = 0;
     pstr_usac_elem_config->usac_ext_ele_payload_present = 0;
     pstr_asc_usac_config->num_elements++;
+    ptr_usac_config->num_ext_elements++;
+    ptr_usac_config->num_elements++;
   }
 
   if (ptr_usac_config->codec_mode == USAC_SWITCHED) {
@@ -1045,7 +1068,7 @@ IA_ERRORCODE ixheaace_usac_encode(FLOAT32 **ptr_input,
   WORD32 len_next_high_rate;
   WORD8 elem_idx, nr_core_coder_channels = 0, chn = 0;
   WORD32 ch_offset = 0;
-  WORD32 elem_idx_max = ptr_usac_config->num_elements;
+  WORD32 elem_idx_max = ptr_usac_config->num_elements - ptr_usac_config->num_ext_elements;
   WORD32 td_buffer_offset = (TD_BUFFER_OFFSET * ptr_usac_config->ccfl) / FRAME_LEN_LONG;
   usac_independency_flg = ptr_usac_data->usac_independency_flag;
 
@@ -1063,15 +1086,16 @@ IA_ERRORCODE ixheaace_usac_encode(FLOAT32 **ptr_input,
     ptr_usac_data->min_bits_needed = 0;
   }
 
-  if (ptr_usac_config->use_drc_element == 1) {
-    elem_idx_max -= 1;
-  }
-
   num_bits = 0;
-
-  iusace_write_bits_buf(pstr_it_bit_buff, usac_independency_flg, 1);
-  num_bits++;
-
+  if (ptr_usac_config->preroll_flag) {
+    if (ptr_usac_config->iframes_interval != ptr_usac_config->num_preroll_frames) {
+      iusace_write_bits_buf(pstr_it_bit_buff, usac_independency_flg, 1);
+      num_bits++;
+    }
+  } else {
+    iusace_write_bits_buf(pstr_it_bit_buff, usac_independency_flg, 1);
+    num_bits++;
+  }
   for (elem_idx = 0; elem_idx < elem_idx_max; elem_idx++) {
     switch (ptr_usac_data->channel_elem_type[elem_idx]) {
       case USAC_ELEMENT_TYPE_SCE:
@@ -1289,7 +1313,12 @@ IA_ERRORCODE ixheaace_usac_encode(FLOAT32 **ptr_input,
         }
       }
     }
-
+    if (ptr_usac_config->preroll_flag) {
+      if (ptr_usac_config->iframes_interval != ptr_usac_config->num_preroll_frames) {
+        iusace_write_bits_buf(pstr_it_bit_buff, 0, 1);  // extension element present
+        num_bits++;
+      }
+    }
     if (ptr_usac_config->use_drc_element) {
       WORD32 num_bits_ext_elem = 0;
       err = iusace_enc_ext_elemts(ID_EXT_ELE_UNI_DRC, ptr_usac_config, pstr_state, pstr_asc,
@@ -1424,4 +1453,4 @@ IA_ERRORCODE ixheaace_usac_encode(FLOAT32 **ptr_input,
   }
 
   return err;
-}
+}

encoder/iusace_main.h

@@ -97,6 +97,9 @@ typedef struct {
   WORD32 min_bits_needed;
   WORD32 num_drc_bits;
   WORD32 use_acelp_only;
+  WORD32 prev_out_bytes[MAX_PREROLL_FRAMES];
+  UWORD8 prev_out_data[MAX_PREROLL_FRAMES][MAX_OUTPUT_BYTES_PER_CH * IXHEAACE_MAX_CH_IN_BS_ELE];
+  UWORD32 stereo_config_index;
 } ia_usac_data_struct;
 
 typedef struct {

encoder/iusace_psy_utils.c

@@ -160,6 +160,9 @@ static VOID iusace_sfb_init(WORD32 sample_rate, WORD32 block_type, WORD32 *ptr_s
       case 8000:
         ptr_sfb_params = pstr_sfb_info_tbls[0].cb_offset_long;
         break;
+      default:
+        ptr_sfb_params = pstr_sfb_info_tbls[8].cb_offset_long;
+        break;
     }
   } else {
     block_len = ccfl >> 3;
@@ -202,6 +205,9 @@ static VOID iusace_sfb_init(WORD32 sample_rate, WORD32 block_type, WORD32 *ptr_s
       case 8000:
         ptr_sfb_params = pstr_sfb_info_tbls[0].cb_offset_short;
         break;
+      default:
+        ptr_sfb_params = pstr_sfb_info_tbls[8].cb_offset_short;
+        break;
     }
   }
 

encoder/iusace_write_bitstream.c

@@ -509,7 +509,7 @@ WORD32 iusace_write_cpe(ia_sfb_params_struct *pstr_sfb_prms, ia_bit_buf_struct *
                              pstr_sfb_prms->num_window_groups[ch], pstr_sfb_prms->max_sfb_ste);
 
     {
-      if (ms_mask == 3) {
+      if ((ms_mask == 3) && (pstr_usac_data->stereo_config_index == 0)) {
         bit_count += iusace_write_cplx_pred_data(
             it_bit_buf, pstr_sfb_prms->num_window_groups[ch], pstr_sfb_prms->max_sfb_ste,
             pstr_usac_data->complex_coef[ch], pstr_usac_data->pred_coef_re[ch],