Merge branch 'feat/add_mn6_ac' into 'master'

Feat/add mn6 ac See merge request speech-recognition-framework/esp-sr!32
2025-09-15 15:28:44 +08:00 · 2023-03-31 16:27:31 +08:00 · 2023-03-31 16:27:31 +08:00 · 1f3692bbd9
commit 1f3692bbd9
parent 629a4f16a6 35cba231ef
20 changed files with 437 additions and 8 deletions
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@ -4,8 +4,9 @@
 - Available storage is less than the remaining flash space on IDF v5.0.   
 If you can not map model partition successfully, please check the left free storage by `spi_flash_mmap_get_free_pages(ESP_PARTITION_MMAP_DATA)`

-## Unreleased
+## 1.3.0 
 - Update the partition APIs to keep compatible with both IDF v4.4 and IDF v5.0
+- Add a new Chinese MultiNet6 mdoel for air conditioner controller 

 ## 1.2.1
 - Fix bugs in model loader
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@ -13,9 +13,12 @@ if(IDF_TARGET STREQUAL "esp32")
    set(requires
        json
        spiffs
-        esp_partition
        )

+    IF (IDF_VERSION_MAJOR GREATER 4)
+        list(APPEND requires esp_partition)
+    ENDIF (IDF_VERSION_MAJOR GREATER 4)
+
    idf_component_register(SRCS ${srcs}
                       INCLUDE_DIRS ${include_dirs}
                       REQUIRES ${requires}
--- a/Kconfig.projbuild
+++ b/Kconfig.projbuild
@ -147,7 +147,11 @@ choice CHINESE_SR_MN_MODEL_SEL
        depends on IDF_TARGET_ESP32S3
    
    config SR_MN_CN_MULTINET6_QUANT
-        bool "chinese recognition (mn6_cn)"
+        bool "general chinese recognition (mn6_cn)"
+        depends on IDF_TARGET_ESP32S3
+    
+    config SR_MN_CN_MULTINET6_AC_QUANT
+        bool "chinese recognition for air conditioner controller (mn6_cn_ac)"
        depends on IDF_TARGET_ESP32S3
 endchoice

@ -166,7 +170,7 @@ choice ENGLISH_SR_MN_MODEL_SEL
        depends on IDF_TARGET_ESP32S3
    
    config SR_MN_EN_MULTINET6_QUANT
-        bool "english recognition (mn6_en)"
+        bool "general english recognition (mn6_en)"
        depends on IDF_TARGET_ESP32S3

 endchoice
--- a/docs/zh_CN/benchmark/README.rst
+++ b/docs/zh_CN/benchmark/README.rst
@ -170,10 +170,10 @@ Word Error Rate 性能测试

 .. note::

-    中文使用没有声调的拼音单元去计算WER.
+    中文使用没有声调的拼音单元去计算WER。

-Speech Commands 性能测试
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~
+Speech Commands 性能测试(空调控制场景)
+~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

 +-----------+-----------+----------+------------+-------------+
 | Model     | Distance  | Quiet    | Stationary | Speech      |
@ -187,6 +187,14 @@ Speech Commands 性能测试
 | MultiNet  | 3 m       |  98.8%   |   88.3%    |   88.0%     |
 | 6_cn      |           |          |            |             |
 +-----------+-----------+----------+------------+-------------+
+| MultiNet  | 3 m       |  97.1%   |   95.1%    |   96.8%     |
+| 6_cn_ac   |           |          |            |             |
+-----------+-----------+----------+------------+-------------+
+
+.. note::
+
+    MultiNet6_cn_ac在空调场景数据集上进行了进一步的微调，所以在空调控制场景具有更好的性能。
+

 TTS
 ---
--- a/idf_component.yml
+++ b/idf_component.yml
@ -1,4 +1,4 @@
-version: "1.2.1"
+version: "1.3.0"
 description: esp_sr provides basic algorithms for Speech Recognition applications
 url: https://github.com/espressif/esp-sr
 dependencies:
--- a/lib/esp32s3/libc_speech_features.a
+++ b/lib/esp32s3/libc_speech_features.a
--- a/lib/esp32s3/libdl_lib.a
+++ b/lib/esp32s3/libdl_lib.a
--- a/lib/esp32s3/libesp_audio_front_end.a
+++ b/lib/esp32s3/libesp_audio_front_end.a
--- a/lib/esp32s3/libesp_audio_processor.a
+++ b/lib/esp32s3/libesp_audio_processor.a
--- a/lib/esp32s3/libfst.a
+++ b/lib/esp32s3/libfst.a
--- a/lib/esp32s3/libhufzip.a
+++ b/lib/esp32s3/libhufzip.a
--- a/lib/esp32s3/libmultinet.a
+++ b/lib/esp32s3/libmultinet.a
--- a/lib/esp32s3/libwakenet.a
+++ b/lib/esp32s3/libwakenet.a
--- a/model/movemodel.py
+++ b/model/movemodel.py
@ -75,6 +75,8 @@ def copy_multinet_from_sdkconfig(model_path, sdkconfig_path, target_path):
        models.append('mn5q8_cn')
    elif "CONFIG_SR_MN_CN_MULTINET6_QUANT" in models_string:
        models.append('mn6_cn')
+    elif "CONFIG_SR_MN_CN_MULTINET6_AC_QUANT" in models_string:
+        models.append('mn6_cn_ac')
    
    if "CONFIG_SR_MN_EN_MULTINET5_SINGLE_RECOGNITION_QUANT8" in models_string:
        models.append('mn5q8_en')
@ -127,6 +129,7 @@ if __name__ == '__main__':

    copy_multinet_from_sdkconfig(model_path, sdkconfig_path, target_path)
    copy_wakenet_from_sdkconfig(model_path, sdkconfig_path, target_path)
+    copy_nsnet_from_sdkconfig(model_path, sdkconfig_path, target_path)
    pack_models(target_path, image_file)
    total_size = os.path.getsize(os.path.join(target_path, image_file))
    recommended_size = int(math.ceil(total_size/1024))
--- a/model/multinet_model/fst/mn6_cn.vocab
+++ b/model/multinet_model/fst/mn6_cn.vocab
--- a/model/multinet_model/mn6_cn_ac/_MODEL_INFO_
+++ b/model/multinet_model/mn6_cn_ac/_MODEL_INFO_
@ -0,0 +1,2 @@
+# (neural network type)_(model data version)_(lable1_detection windown length_threshold for 90%_threshold for 95%)_(lable2 ...)_...
+MN6_AC_chinese_8_0.9_0.90
--- a/model/multinet_model/mn6_cn_ac/mn6_data
+++ b/model/multinet_model/mn6_cn_ac/mn6_data
--- a/model/multinet_model/mn6_cn_ac/mn6_index
+++ b/model/multinet_model/mn6_cn_ac/mn6_index
--- a/model/multinet_model/mn6_cn_ac/vocab
+++ b/model/multinet_model/mn6_cn_ac/vocab
@ -0,0 +1,408 @@
+<blk>	0
+<sos/eos>	0
+<unk>	0
+▁a	-6.00389
+▁ai	-6.00389
+▁an	-6.00389
+▁ang	-6.00389
+▁ao	-6.00389
+▁ba	-6.00389
+▁bai	-6.00389
+▁ban	-6.00389
+▁bang	-6.00389
+▁bao	-6.00389
+▁bei	-6.00389
+▁ben	-6.00389
+▁beng	-6.00389
+▁bi	-6.00389
+▁bian	-6.00389
+▁biao	-6.00389
+▁bie	-6.00389
+▁bin	-6.00389
+▁bing	-6.00389
+▁bo	-6.00389
+▁bu	-6.00389
+▁ca	-6.00389
+▁cai	-6.00389
+▁can	-6.00389
+▁cang	-6.00389
+▁cao	-6.00389
+▁ce	-6.00389
+▁cen	-6.00389
+▁ceng	-6.00389
+▁cha	-6.00389
+▁chai	-6.00389
+▁chan	-6.00389
+▁chang	-6.00389
+▁chao	-6.00389
+▁che	-6.00389
+▁chen	-6.00389
+▁cheng	-6.00389
+▁chi	-6.00389
+▁chong	-6.00389
+▁chou	-6.00389
+▁chu	-6.00389
+▁chuai	-6.00389
+▁chuan	-6.00389
+▁chuang	-6.00389
+▁chui	-6.00389
+▁chun	-6.00389
+▁chuo	-6.00389
+▁ci	-6.00389
+▁cong	-6.00389
+▁cou	-6.00389
+▁cu	-6.00389
+▁cuan	-6.00389
+▁cui	-6.00389
+▁cun	-6.00389
+▁cuo	-6.00389
+▁da	-6.00389
+▁dai	-6.00389
+▁dan	-6.00389
+▁dang	-6.00389
+▁dao	-6.00389
+▁de	-6.00389
+▁dei	-6.00389
+▁deng	-6.00389
+▁di	-6.00389
+▁dia	-6.00389
+▁dian	-6.00389
+▁diao	-6.00389
+▁die	-6.00389
+▁ding	-6.00389
+▁diu	-6.00389
+▁dong	-6.00389
+▁dou	-6.00389
+▁du	-6.00389
+▁duan	-6.00389
+▁dui	-6.00389
+▁dun	-6.00389
+▁duo	-6.00389
+▁e	-6.00389
+▁ei	-6.00389
+▁en	-6.00389
+▁eng	-6.00389
+▁er	-6.00389
+▁fa	-6.00389
+▁fan	-6.00389
+▁fang	-6.00389
+▁fei	-6.00389
+▁fen	-6.00389
+▁feng	-6.00389
+▁fo	-6.00389
+▁fou	-6.00389
+▁fu	-6.00389
+▁ga	-6.00389
+▁gai	-6.00389
+▁gan	-6.00389
+▁gang	-6.00389
+▁gao	-6.00389
+▁ge	-6.00389
+▁gei	-6.00389
+▁gen	-6.00389
+▁geng	-6.00389
+▁gong	-6.00389
+▁gou	-6.00389
+▁gu	-6.00389
+▁gua	-6.00389
+▁guai	-6.00389
+▁guan	-6.00389
+▁guang	-6.00389
+▁gui	-6.00389
+▁gun	-6.00389
+▁guo	-6.00389
+▁ha	-6.00389
+▁hai	-6.00389
+▁han	-6.00389
+▁hang	-6.00389
+▁hao	-6.00389
+▁he	-6.00389
+▁hei	-6.00389
+▁hen	-6.00389
+▁heng	-6.00389
+▁hong	-6.00389
+▁hou	-6.00389
+▁hu	-6.00389
+▁hua	-6.00389
+▁huai	-6.00389
+▁huan	-6.00389
+▁huang	-6.00389
+▁hui	-6.00389
+▁hun	-6.00389
+▁huo	-6.00389
+▁ji	-6.00389
+▁jia	-6.00389
+▁jian	-6.00389
+▁jiang	-6.00389
+▁jiao	-6.00389
+▁jie	-6.00389
+▁jin	-6.00389
+▁jing	-6.00389
+▁jiong	-6.00389
+▁jiu	-6.00389
+▁ju	-6.00389
+▁juan	-6.00389
+▁jue	-6.00389
+▁jun	-6.00389
+▁ka	-6.00389
+▁kai	-6.00389
+▁kan	-6.00389
+▁kang	-6.00389
+▁kao	-6.00389
+▁ke	-6.00389
+▁ken	-6.00389
+▁keng	-6.00389
+▁kong	-6.00389
+▁kou	-6.00389
+▁ku	-6.00389
+▁kua	-6.00389
+▁kuai	-6.00389
+▁kuan	-6.00389
+▁kuang	-6.00389
+▁kui	-6.00389
+▁kun	-6.00389
+▁kuo	-6.00389
+▁la	-6.00389
+▁lai	-6.00389
+▁lan	-6.00389
+▁lang	-6.00389
+▁lao	-6.00389
+▁le	-6.00389
+▁lei	-6.00389
+▁leng	-6.00389
+▁li	-6.00389
+▁lia	-6.00389
+▁lian	-6.00389
+▁liang	-6.00389
+▁liao	-6.00389
+▁lie	-6.00389
+▁lin	-6.00389
+▁ling	-6.00389
+▁liu	-6.00389
+▁long	-6.00389
+▁lou	-6.00389
+▁lu	-6.00389
+▁luan	-6.00389
+▁lun	-6.00389
+▁luo	-6.00389
+▁lv	-6.00389
+▁lve	-6.00389
+▁ma	-6.00389
+▁mai	-6.00389
+▁man	-6.00389
+▁mang	-6.00389
+▁mao	-6.00389
+▁me	-6.00389
+▁mei	-6.00389
+▁men	-6.00389
+▁meng	-6.00389
+▁mi	-6.00389
+▁mian	-6.00389
+▁miao	-6.00389
+▁mie	-6.00389
+▁min	-6.00389
+▁ming	-6.00389
+▁miu	-6.00389
+▁mo	-6.00389
+▁mou	-6.00389
+▁mu	-6.00389
+▁n	-6.00389
+▁na	-6.00389
+▁nai	-6.00389
+▁nan	-6.00389
+▁nang	-6.00389
+▁nao	-6.00389
+▁ne	-6.00389
+▁nei	-6.00389
+▁nen	-6.00389
+▁neng	-6.00389
+▁ni	-6.00389
+▁nian	-6.00389
+▁niang	-6.00389
+▁niao	-6.00389
+▁nie	-6.00389
+▁nin	-6.00389
+▁ning	-6.00389
+▁niu	-6.00389
+▁nong	-6.00389
+▁nou	-6.00389
+▁nu	-6.00389
+▁nuan	-6.00389
+▁nuo	-6.00389
+▁nv	-6.00389
+▁nve	-6.00389
+▁o	-6.00389
+▁ou	-6.00389
+▁pa	-6.00389
+▁pai	-6.00389
+▁pan	-6.00389
+▁pang	-6.00389
+▁pao	-6.00389
+▁pei	-6.00389
+▁pen	-6.00389
+▁peng	-6.00389
+▁pi	-6.00389
+▁pian	-6.00389
+▁piao	-6.00389
+▁pie	-6.00389
+▁pin	-6.00389
+▁ping	-6.00389
+▁po	-6.00389
+▁pou	-6.00389
+▁pu	-6.00389
+▁qi	-6.00389
+▁qia	-6.00389
+▁qian	-6.00389
+▁qiang	-6.00389
+▁qiao	-6.00389
+▁qie	-6.00389
+▁qin	-6.00389
+▁qing	-6.00389
+▁qiong	-6.00389
+▁qiu	-6.00389
+▁qu	-6.00389
+▁quan	-6.00389
+▁que	-6.00389
+▁qun	-6.00389
+▁ran	-6.00389
+▁rang	-6.00389
+▁rao	-6.00389
+▁re	-6.00389
+▁ren	-6.00389
+▁reng	-6.00389
+▁ri	-6.00389
+▁rong	-6.00389
+▁rou	-6.00389
+▁ru	-6.00389
+▁ruan	-6.00389
+▁rui	-6.00389
+▁run	-6.00389
+▁ruo	-6.00389
+▁sa	-6.00389
+▁sai	-6.00389
+▁san	-6.00389
+▁sang	-6.00389
+▁sao	-6.00389
+▁se	-6.00389
+▁sen	-6.00389
+▁seng	-6.00389
+▁sha	-6.00389
+▁shai	-6.00389
+▁shan	-6.00389
+▁shang	-6.00389
+▁shao	-6.00389
+▁she	-6.00389
+▁shei	-6.00389
+▁shen	-6.00389
+▁sheng	-6.00389
+▁shi	-6.00389
+▁shou	-6.00389
+▁shu	-6.00389
+▁shua	-6.00389
+▁shuai	-6.00389
+▁shuan	-6.00389
+▁shuang	-6.00389
+▁shui	-6.00389
+▁shun	-6.00389
+▁shuo	-6.00389
+▁si	-6.00389
+▁song	-6.00389
+▁sou	-6.00389
+▁su	-6.00389
+▁suan	-6.00389
+▁sui	-6.00389
+▁sun	-6.00389
+▁suo	-6.00389
+▁ta	-6.00389
+▁tai	-6.00389
+▁tan	-6.00389
+▁tang	-6.00389
+▁tao	-6.00389
+▁te	-6.00389
+▁teng	-6.00389
+▁ti	-6.00389
+▁tian	-6.00389
+▁tiao	-6.00389
+▁tie	-6.00389
+▁ting	-6.00389
+▁tong	-6.00389
+▁tou	-6.00389
+▁tu	-6.00389
+▁tuan	-6.00389
+▁tui	-6.00389
+▁tun	-6.00389
+▁tuo	-6.00389
+▁wa	-6.00389
+▁wai	-6.00389
+▁wan	-6.00389
+▁wang	-6.00389
+▁wei	-6.00389
+▁wen	-6.00389
+▁weng	-6.00389
+▁wo	-6.00389
+▁wu	-6.00389
+▁xi	-6.00389
+▁xia	-6.00389
+▁xian	-6.00389
+▁xiang	-6.00389
+▁xiao	-6.00389
+▁xie	-6.00389
+▁xin	-6.00389
+▁xing	-6.00389
+▁xiong	-6.00389
+▁xiu	-6.00389
+▁xu	-6.00389
+▁xuan	-6.00389
+▁xue	-6.00389
+▁xun	-6.00389
+▁ya	-6.00389
+▁yan	-6.00389
+▁yang	-6.00389
+▁yao	-6.00389
+▁ye	-6.00389
+▁yi	-6.00389
+▁yin	-6.00389
+▁ying	-6.00389
+▁yo	-6.00389
+▁yong	-6.00389
+▁you	-6.00389
+▁yu	-6.00389
+▁yuan	-6.00389
+▁yue	-6.00389
+▁yun	-6.00389
+▁za	-6.00389
+▁zai	-6.00389
+▁zan	-6.00389
+▁zang	-6.00389
+▁zao	-6.00389
+▁ze	-6.00389
+▁zei	-6.00389
+▁zen	-6.00389
+▁zeng	-6.00389
+▁zha	-6.00389
+▁zhai	-6.00389
+▁zhan	-6.00389
+▁zhang	-6.00389
+▁zhao	-6.00389
+▁zhe	-6.00389
+▁zhen	-6.00389
+▁zheng	-6.00389
+▁zhi	-6.00389
+▁zhong	-6.00389
+▁zhou	-6.00389
+▁zhu	-6.00389
+▁zhua	-6.00389
+▁zhuai	-6.00389
+▁zhuan	-6.00389
+▁zhuang	-6.00389
+▁zhui	-6.00389
+▁zhun	-6.00389
+▁zhuo	-6.00389
+▁zi	-6.00389
+▁zong	-6.00389
+▁zou	-6.00389
+▁zu	-6.00389
+▁zuan	-6.00389
+▁zui	-6.00389
+▁zun	-6.00389
+▁zuo	-6.00389
--- a/model/multinet_model/fst/mn6_en.vocab
+++ b/model/multinet_model/fst/mn6_en.vocab