Merge branch 'feat/s3_wn9s' into 'master'

feat: esp32,esp32s3,esp32p4 support wakenet9s

See merge request speech-recognition-framework/esp-sr!176

Kconfig.projbuild

@@ -54,7 +54,6 @@ endchoice
 
 
 menu "Load Multiple Wake Words (WakeNet9s)"
-    depends on IDF_TARGET_ESP32C5 ||  IDF_TARGET_ESP32C3 || IDF_TARGET_ESP32C6 || IDF_TARGET_ESP32S2
 
     config SR_WN_WN9S_HILEXIN
     bool "Hi,乐鑫 (wn9s_hilexin)"

include/esp32/esp_mfcc_iface.h

@@ -68,6 +68,8 @@ typedef esp_mfcc_data_t *(*esp_mfcc_op_create_t)(const esp_mfcc_opts_t *opt);
  */
 typedef float *(*esp_mfcc_op_run_step_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t nch);
 
+typedef void (*esp_mfcc_op_run_step_s16_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t *fbank);
+
 /**
  * @brief Clean all state of mfcc handle
  *
@@ -82,5 +84,6 @@ typedef struct {
     esp_mfcc_op_destroy_t destroy;
     esp_mfcc_op_create_t create;
     esp_mfcc_op_run_step_t run_step;
+    esp_mfcc_op_run_step_s16_t run_step_s16;
     esp_mfcc_op_clean_t clean;
 } esp_mfcc_iface_t;

include/esp32/esp_mfcc_models.h

@@ -2,12 +2,18 @@
 #include "esp_mfcc_iface.h"
 
 extern const esp_mfcc_iface_t esp_fbank_f32; // float32-fbank handle
+extern const esp_mfcc_iface_t esp_fbank_s16; // int16-fbank handle
 
 /**
  * @brief Return basic opts used in wakenet9 & multinet5
  **/
 esp_mfcc_opts_t *get_mfcc_opts_wn9();
 
+/**
+ * @brief Return basic opts used in wakenet9s
+ **/
+esp_mfcc_opts_t *get_mfcc_opts_wn9s16();
+
 /**
  * @brief Return basic opts for default kaldifeat
  *

include/esp32/esp_speech_features.h

@@ -48,7 +48,7 @@ float *esp_win_func_init(char *win_type, float *window_data, int frame_length);
 
 float *esp_fftr(float *x, int nfft, void *fft_table);
 
-float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_table);
+float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_handle);
 
 void esp_audio_short_to_float(short *samples, float *x, int len, int remove_dc);
 

include/esp32/esp_wn_iface.h

@@ -29,6 +29,7 @@ typedef enum {
     DET_MODE_2CH_95 = 3,
     DET_MODE_3CH_90 = 4,
     DET_MODE_3CH_95 = 5,
+	DET_MODE_90_COPY_PARAMS = 6,       // Aggressive
 } det_mode_t;
 
 typedef struct {

include/esp32p4/esp_mfcc_iface.h

@@ -68,6 +68,8 @@ typedef esp_mfcc_data_t *(*esp_mfcc_op_create_t)(const esp_mfcc_opts_t *opt);
  */
 typedef float *(*esp_mfcc_op_run_step_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t nch);
 
+typedef void (*esp_mfcc_op_run_step_s16_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t *fbank);
+
 /**
  * @brief Clean all state of mfcc handle
  *
@@ -82,5 +84,6 @@ typedef struct {
     esp_mfcc_op_destroy_t destroy;
     esp_mfcc_op_create_t create;
     esp_mfcc_op_run_step_t run_step;
+    esp_mfcc_op_run_step_s16_t run_step_s16;
     esp_mfcc_op_clean_t clean;
 } esp_mfcc_iface_t;

include/esp32p4/esp_mfcc_models.h

@@ -2,12 +2,18 @@
 #include "esp_mfcc_iface.h"
 
 extern const esp_mfcc_iface_t esp_fbank_f32; // float32-fbank handle
+extern const esp_mfcc_iface_t esp_fbank_s16; // int16-fbank handle
 
 /**
  * @brief Return basic opts used in wakenet9 & multinet5
  **/
 esp_mfcc_opts_t *get_mfcc_opts_wn9();
 
+/**
+ * @brief Return basic opts used in wakenet9s
+ **/
+esp_mfcc_opts_t *get_mfcc_opts_wn9s16();
+
 /**
  * @brief Return basic opts for default kaldifeat
  *

include/esp32p4/esp_speech_features.h

@@ -48,7 +48,7 @@ float *esp_win_func_init(char *win_type, float *window_data, int frame_length);
 
 float *esp_fftr(float *x, int nfft, void *fft_table);
 
-float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_table);
+float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_handle);
 
 void esp_audio_short_to_float(short *samples, float *x, int len, int remove_dc);
 

include/esp32p4/esp_wn_iface.h

@@ -29,6 +29,7 @@ typedef enum {
     DET_MODE_2CH_95 = 3,
     DET_MODE_3CH_90 = 4,
     DET_MODE_3CH_95 = 5,
+	DET_MODE_90_COPY_PARAMS = 6,       // Aggressive
 } det_mode_t;
 
 typedef struct {

include/esp32s3/esp_mfcc_iface.h

@@ -68,6 +68,8 @@ typedef esp_mfcc_data_t *(*esp_mfcc_op_create_t)(const esp_mfcc_opts_t *opt);
  */
 typedef float *(*esp_mfcc_op_run_step_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t nch);
 
+typedef void (*esp_mfcc_op_run_step_s16_t)(esp_mfcc_data_t *r, int16_t *samp, int16_t *fbank);
+
 /**
  * @brief Clean all state of mfcc handle
  *
@@ -82,5 +84,6 @@ typedef struct {
     esp_mfcc_op_destroy_t destroy;
     esp_mfcc_op_create_t create;
     esp_mfcc_op_run_step_t run_step;
+    esp_mfcc_op_run_step_s16_t run_step_s16;
     esp_mfcc_op_clean_t clean;
 } esp_mfcc_iface_t;

include/esp32s3/esp_mfcc_models.h

@@ -2,12 +2,18 @@
 #include "esp_mfcc_iface.h"
 
 extern const esp_mfcc_iface_t esp_fbank_f32; // float32-fbank handle
+extern const esp_mfcc_iface_t esp_fbank_s16; // int16-fbank handle
 
 /**
  * @brief Return basic opts used in wakenet9 & multinet5
  **/
 esp_mfcc_opts_t *get_mfcc_opts_wn9();
 
+/**
+ * @brief Return basic opts used in wakenet9s
+ **/
+esp_mfcc_opts_t *get_mfcc_opts_wn9s16();
+
 /**
  * @brief Return basic opts for default kaldifeat
  *

include/esp32s3/esp_speech_features.h

@@ -48,7 +48,7 @@ float *esp_win_func_init(char *win_type, float *window_data, int frame_length);
 
 float *esp_fftr(float *x, int nfft, void *fft_table);
 
-float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_table);
+float *esp_spectrum_step(float *x, int nfft, bool use_power, void *fft_handle);
 
 void esp_audio_short_to_float(short *samples, float *x, int len, int remove_dc);
 

include/esp32s3/esp_wn_iface.h

@@ -29,6 +29,7 @@ typedef enum {
     DET_MODE_2CH_95 = 3,
     DET_MODE_3CH_90 = 4,
     DET_MODE_3CH_95 = 5,
+	DET_MODE_90_COPY_PARAMS = 6,       // Aggressive
 } det_mode_t;
 
 typedef struct {

test_apps/esp-sr/sdkconfig.ci.wn9s_hilexin

@@ -0,0 +1,25 @@
+# This file was generated using idf.py save-defconfig. It can be edited manually.
+# Espressif IoT Development Framework (ESP-IDF) 5.5.0 Project Minimal Configuration
+#
+CONFIG_IDF_TARGET="esp32s3"
+CONFIG_ESPTOOLPY_FLASHMODE_QIO=y
+CONFIG_ESPTOOLPY_FLASHSIZE_16MB=y
+CONFIG_PARTITION_TABLE_CUSTOM=y
+CONFIG_SR_VADN_VADNET1_MEDIUM=y
+CONFIG_SR_WN_WN9S_HILEXIN=y
+CONFIG_SPIRAM=y
+CONFIG_ESP_TASK_WDT_EN=n
+CONFIG_ESP_TASK_WDT_INIT=n
+CONFIG_SPIRAM_MODE_OCT=y
+CONFIG_SPIRAM_SPEED_80M=y
+CONFIG_ESP_DEFAULT_CPU_FREQ_MHZ_240=y
+CONFIG_ESP32S3_INSTRUCTION_CACHE_32KB=y
+CONFIG_ESP32S3_DATA_CACHE_64KB=y
+CONFIG_ESP32S3_DATA_CACHE_LINE_64B=y
+CONFIG_ESP_MAIN_TASK_STACK_SIZE=8192
+CONFIG_ESP_WIFI_GMAC_SUPPORT=n
+CONFIG_FREERTOS_VTASKLIST_INCLUDE_COREID=y
+CONFIG_FREERTOS_GENERATE_RUN_TIME_STATS=y
+CONFIG_LWIP_TCP_SND_BUF_DEFAULT=5744
+CONFIG_LWIP_TCP_WND_DEFAULT=5744
+CONFIG_UNITY_CRITICAL_LEAK_LEVEL_GENERAL=1024