Merge branch 'feat/add_nsnet2_model' into 'master'

feat/add_nsnet2

See merge request speech-recognition-framework/esp-sr!93

Kconfig.projbuild

@@ -46,7 +46,9 @@ choice SR_NSN_MODEL_LOAD
     config SR_NSN_NSNET1
         bool "Deep noise suppression v1 (nsnet1)"
         depends on IDF_TARGET_ESP32S3
-
+    config SR_NSN_NSNET2
+        bool "Deep noise suppression v2 (nsnet2)"
+	depends on IDF_TARGET_ESP32S3
 endchoice
 
 config USE_WAKENET
@@ -136,6 +138,10 @@ choice SR_WN_MODEL_LOAD
         bool "Hi,Xiaoxing (wn9_hixiaoxing_tts)"
         depends on IDF_TARGET_ESP32S3
 
+    config SR_WN_WN9_MYCROFT_TTS
+        bool "Mycroft (wn9_mycroft_tts)"
+        depends on IDF_TARGET_ESP32S3
+
     config SR_WN_WN9_CUSTOMWORD
         bool "customized word (wn9_customword)"
         depends on IDF_TARGET_ESP32S3
@@ -189,6 +195,10 @@ menu "Load Multiple Wake Words"
     bool "xiaomeitongxue (wn9_xiaomeitongxue_tts)"
     depends on IDF_TARGET_ESP32S3
 
+    config SR_WN_WN9_MYCROFT_TTS_MULTI
+    bool "Mycroft (wn9_mycroft_tts)"
+    depends on IDF_TARGET_ESP32S3
+
 endmenu
 
 config USE_MULTINET

README.md

@@ -41,6 +41,7 @@ The following wake words are supported in esp-sr:
 |Computer         |                                        | wn9_computer_tts       |
 |Hey,Willow       |                                        | wn9_heywillow_tts      | 
 |Sophia           |                                        | wn9_sophia_tts         |
+|Mycroft          |                                        | wn9_mycroft_tts        |
 |你好小鑫         |                                        | wn9_nihaoxiaoxin_tts   |
 |小美同学         |                                        | wn9_xiaomeitongxue_tts |
 |Hi,小星          |                                        | wn9_hixiaoxing_tts     |

include/esp32s3/dl_lib.h

@@ -78,7 +78,7 @@ void *dl_lib_calloc_psram(int cnt, int size, int align);
 /**
  * @brief Free aligned memory allocated by `dl_lib_calloc` or `dl_lib_calloc_psram` 
  * 
- * @param prt    Pointer to free
+ * @param ptr    Pointer to free
  */
 void dl_lib_free(void *ptr);
 
@@ -415,4 +415,4 @@ dl_matrix2d_t *dl_basic_conv_layer_quantised_weight(const dl_matrix2d_t *in, con
 }
 #endif
 
-#endif
+#endif

include/esp32s3/dl_lib_convq8_queue.h

@@ -292,6 +292,7 @@ qtp_t *dl_atrous_conv1dq8_16_s3(dl_convq8_queue_t *in, dl_convq_queue_t *out, in
 
 void print_convq8(dl_convq8_queue_t *cq, int offset);
 void print_convq(dl_convq_queue_t *cq, int offset);
+void dl_relu_convq8(dl_convq8_queue_t *cq);
 
 void lstmq8_free(void);
 

include/esp32s3/dl_lib_convq_queue.h

@@ -279,9 +279,9 @@ dl_matrix2dq_t *dl_convq_lstm_layer(const dl_convq_queue_t *in, dl_convq_queue_t
 dl_matrix2dq_t *dl_basic_lstm_layer1_q(const dl_convq_queue_t *in, dl_matrix2dq_t *state_c, dl_matrix2dq_t *state_h,
                                        const dl_matrix2dq_t *weight, const dl_matrix2dq_t *bias, int step, int shift);
 
-dl_matrix2dq_t *dl_convq16_lstm_layer(const dl_convq_queue_t *in, dl_convq_queue_t *out, dl_matrix2dq_t *state_c,
+dl_matrix2dq_t *dl_convq16_lstm_layer(dl_convq_queue_t *in, dl_convq_queue_t *out, dl_matrix2dq_t *state_c,
-                                       dl_matrix2dq_t *state_h, const dl_matrix2dq_t *in_weight, const dl_matrix2dq_t *h_weight,
+                                       dl_matrix2dq_t *state_h, dl_matrix2dq_t *in_weight, dl_matrix2dq_t *h_weight,
-                                       const dl_matrix2dq_t *bias, int prenum);
+                                       dl_matrix2dq_t *bias, int prenum);
 
 /**
  * @brief Allocate a fixed-point multi channel convolution queue 

include/esp32s3/esp_nsn_models.h

@@ -2,8 +2,16 @@
 
 #include "esp_nsn_iface.h"
 
-// The prefix of nsnet model name is used to filter all wakenet from availabel models.
+/*
+The prefix of nset
+Now there are nsnet1 and nsnet2
+*/
 #define ESP_NSNET_PREFIX "nsnet"
 
-extern const esp_nsn_iface_t esp_nsnet1_quantized;
+/**
-#define ESP_NSN_HANDLE esp_nsnet1_quantized
+ * @brief Get the nsnet handle from model name
+ *
+ * @param model_name   The name of model 
+ * @returns The handle of multinet
+ */
+esp_nsn_iface_t *esp_nsnet_handle_from_name(char *model_name);

model/movemodel.py

@@ -97,6 +97,8 @@ def copy_nsnet_from_sdkconfig(model_path, sdkconfig_path, target_path):
     models = []
     if "CONFIG_SR_NSN_NSNET1" in models_string:
         models.append('nsnet1')
+    if "CONFIG_SR_NSN_NSNET2" in models_string:
+        models.append('nsnet2')
 
     for item in models:
         shutil.copytree(model_path + '/nsnet_model/' + item, target_path+'/'+item)

model/nsnet_model/nsnet2/_MODEL_INFO_

@@ -0,0 +1,2 @@
+# (neural network type)_(model data version)_ns_0_0.0_0.0
+nsnet2_v1_ns_0_0.0_0.0

model/wakenet_model/wn9_mycroft_tts/_MODEL_INFO_

@@ -0,0 +1 @@
+wakenet9l_tts1h8_Mycroft_3_0.625_0.629

test_apps/esp-sr/pytest_esp_sr.py

@@ -57,10 +57,11 @@ def test_sr_afe(dut: Dut)-> None:
     'config',
     [
         'nsnet1',
+        'nsnet2',
     ],
 )
 def test_vc_afe(dut: Dut)-> None:
     # dut.run_all_single_board_cases(group="afe")
     dut.expect_exact('Press ENTER to see the list of tests.')
     dut.write('[afe_vc]')
-    dut.expect_unity_test_output(timeout = 100000)
+    dut.expect_unity_test_output(timeout = 100000)