Speech Recognition | Seeed Studio Wiki

[giscus[bot]]

Speech Recognition | Seeed Studio Wiki

Speech Recognition

https://wiki.seeedstudio.com/XIAO-BLE-Sense-TFLite-Mic/

[stanleyoz]

Nice write up! We getting this error after uploading our file over to the Sense board ..

Feature generation failed
Requested feature_data_ size 536871936 doesn't match 1960

Anyone can point to where I probably messed up along the way? Thanks

[MatthewJeffson]

Welcome to our product documentation platform! We value user feedback and contributions, so we couldn’t be more grateful that you can help us improving the documentation.

Hence, if you find obvious errors in the product documentation, or you have additional ideas and insights about using products, please contribute to us.

If you need technical support, you can email it to fae@seeed.cc.

If you want to discuss the product, I recommend the Seeed Studio Forum and Seeed Studio Discord Channel

The "Discussion channel" in Github here is a relatively new addition and may not receive much attention.

Nonetheless, we hope you find it helpful with the help of others. We will clean it up on a monthly basis.

[ahsanfi]

Hello, I'm trying to record audio (.wav) to sd card with XIAO BLE SENSE nRF5840 with seeduino extension, but keep getting failed like this:

Capturing .wav samples
initialization failed!

I think the module cannot detect the sdcard. Any solutions? Thanks

[xzxcessarr]

Hello, Could you tell us how much space of your tf card and test the PDM microphone by the example mic_serial_plotter? This error may caused by the tf card or the PDM microphone, we need futher locate the error by your help

[ahsanfi]

My problem has solved. I think it because the SD Card capacity. I get error when I use 64GB sd card, then I change it to 16 GB. And its work Sent from Outlook for Android<https://aka.ms/AAb9ysg>

…

________________________________ From: xzxcessarr ***@***.***> Sent: Wednesday, August 23, 2023 2:09:11 PM To: Seeed-Studio/wiki-documents ***@***.***> Cc: Muhammad Ahsan Fillah Abadi ***@***.***>; Comment ***@***.***> Subject: Re: [Seeed-Studio/wiki-documents] Speech Recognition | Seeed Studio Wiki (Discussion #192) Hello, Could you tell us how much space of your tf card and test the PDM microphone by the example mic_serial_plotter? This error may caused by the tf card or the PDM microphone, we need futher locate the error by your help — Reply to this email directly, view it on GitHub<#192 (reply in thread)>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AZK353BOAFAEOUIBVNF2WCTXWWUBPANCNFSM6AAAAAAXXZSQFA>. You are receiving this because you commented.Message ID: ***@***.***>

[xzxcessarr]

Thanks for feedback!

[javiBajoCero]

If someone else is getting
"error: conflicting declaration 'unsigned int g_model_len'
unsigned int g_model_len = 19160;"

Is because we blindly copypasted everything including g_model_len at the end of model.cc

The python script is declaring g_model_len as unsigned.

You fix this by removing the unsigned in unsigned int g_model_len = 19160;" inside the cpp file

[javiBajoCero]

Alright that will make it compile but this error will show in the serial monitor

"Feature generation failed"
"Requested feaure_data_size 536871936 doesnt match 1960"

That feels like a signed integer overflow? or i messed something up with the copypasting

[javiBajoCero]

This guy had the same problem and fixed it , tensorflow/tensorflow#61074

[javiBajoCero]

Also this guy tensorflow/tensorflow#44868

[nfs0619]

Thank you for your valuable feedback and for sharing the solution with us! We truly appreciate you taking the time to help improve the experience for everyone. If you encounter any other issues or have suggestions, please don’t hesitate to reach out.

[valentas]

This is not working on PlatformIO

tried to compile example but stuck with some (PDMFilter_CM7_GCC) lib not found.

log at the end:
Archiving .pio/build/stable/libFrameworkArduino.a
Indexing .pio/build/stable/libFrameworkArduino.a
Archiving .pio/build/stable/libed2/libtflite-micro-arduino-examples.a
Indexing .pio/build/stable/libed2/libtflite-micro-arduino-examples.a
Linking .pio/build/stable/firmware.elf
/home/v/.platformio/packages/toolchain-gccarmnoneeabi/bin/../lib/gcc/arm-none-eabi/8.2.1/../../../../arm-none-eabi/bin/ld: cannot find -lPDMFilter_CM7_GCC_wc32
collect2: error: ld returned 1 exit status
*** [.pio/build/stable/firmware.elf] Error 1

platformio.ini:

[env:stable]
platform = nordicnrf52
board = xiaoblesense
framework = arduino
monitor_speed = 115200
build_flags =
-DCORE_DEBUG_LEVEL=3

[valentas]

I also tried suggested platformio.ini config and got same results:

[env]
platform = https://github.com/maxgerhardt/platform-nordicnrf52
framework = arduino
monitor_speed = 115200
build_flags =
-DCORE_DEBUG_LEVEL=3

[env:xiaoblesense_arduinocore_mbed]
board = xiaoblesense

[env:xiaoble_arduinocore_mbed]
board = xiaoble

[nfs0619]

Hello, the compilation error is due to the missing PDMFilter_CM7_GCC library. It seems that the platformio.ini configuration may not correctly reference the customized Nordic nRF52 platform. We kindly suggest checking if the board configuration is correct and ensuring that the missing library files (such as hal_input_pdm_mic.c and PDMFilter_CM7_GCC) are placed in the project's drivers or libs directory. If the issue persists, please feel free to provide more detailed steps and logs to our technical support email at techsupport@seeed.io, and they will be happy to assist you further.

[jasg1612]

After crying so much and trying with that mistake <<Feature generation failed
Requested feature_data_ size 536871936 doesn't match 1960>>, God brought me to a solution. go to the ino file, which according to the library installed is called <<micro_speech.ino>>
step 01
In this line of code the default code is 4 change to 5
static tflite::MicroMutableOpResolver<5> micro_op_resolver(error_reporter);
step 02, in these lines of code
if (micro_op_resolver.AddDepthwiseConv2D() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddFullyConnected() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddSoftmax() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddReshape() != kTfLiteOk) {
return;
}

add the code at the end like this
static tflite::MicroMutableOpResolver<5> micro_op_resolver(error_reporter);
if (micro_op_resolver.AddDepthwiseConv2D() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddFullyConnected() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddSoftmax() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddReshape() != kTfLiteOk) {
return;
}
if (micro_op_resolver.AddConv2D() != kTfLiteOk) {
return;
} // I have added the Conv2D op and deleted the DepthwiseConv2D op.

[nfs0619]

Thanks for sharing!