Using the microphone (MP34DT01) on the STM32 Discovery Board (ST B-L475E-IOT01A)

,
#1

Hey guys,

I’m currently working on trying to replicate the “Faucet Tutorial” from EdgeImpulse but via STM32CubeIDE. I wanted to make this one work first so I can work on a bigger project later on. At the moment I’m having issues trying to make my EdgeImpulse model work with the inputs from the microphone. Here’s what my code looks like for now :

ei_printf(“Init Audio\n”);
AudioInit.BitsPerSample = 16; //16
AudioInit.ChannelsNbr = AUDIO_CHANNELS;
AudioInit.Device = AUDIO_IN_DIGITAL_MIC1; //AUDIO_IN_DIGITAL_MIC1
AudioInit.SampleRate = AUDIO_FREQUENCY_16K; //AUDIO_FREQUENCY_11K // AUDIO_SAMPLING_FREQUENCY //
AudioInit.Volume = 100; //32

int32_t AudioInitRet = BSP_AUDIO_IN_Init(AUDIO_INSTANCE, &AudioInit);

if (AudioInitRet != BSP_ERROR_NONE) {
ei_printf(“Error Audio Init (%ld)\r\n”, AudioInitRet);
}
else {
ei_printf(“OK Audio Init\t(Audio Freq=%ld)\r\n”, AudioInit.SampleRate);
}

/* USER CODE END 2 */

/* Infinite loop /
/ USER CODE BEGIN WHILE /
while (1)
{
/ USER CODE END WHILE */

  int32_t ret2 = BSP_AUDIO_IN_Record(AUDIO_INSTANCE, (uint8_t *) ram_buffer, 2048U); //64U // 2048U // Max 4095U
  if (ret2 != BSP_ERROR_NONE) {
	  ei_printf("Error Audio Record (%ld)\n", ret2);
  }
  else {

	  for(uint32_t ui = 0; ui <= sizeof(ram_buffer); ui++)
	  {
		  ei_printf("%d\t", ram_buffer[ui]);
	  }

	  ei_printf("OK Audio Record\n");

	  signal_t signal;
	  signal.total_length = TOTAL_AUDIO_SIZE; // 16000
	  signal.get_data = &ei_microphone_audio_signal_get_data;
	  ei_impulse_result_t result = { 0 };

	  ei_printf("%d\t", signal.get_data);

	  ei_printf("run_classifier_continuous\n");
	  EI_IMPULSE_ERROR r = run_classifier_continuous(&signal, &result, true);


	  if (r != EI_IMPULSE_OK) {
		  ei_printf("ERR: Failed to run classifier (%d)\n", r);
	  }
	  else {
		  // Print the predictions
		  ei_printf("Predictions (DSP: %d ms., Classification: %d ms., Anomaly: %d ms.):\n",
					result.timing.dsp, result.timing.classification, result.timing.anomaly);
		  for (size_t ix = 0; ix < EI_CLASSIFIER_LABEL_COUNT; ix++) {
			  ei_printf("%s: ", result.classification[ix].label);
			  ei_printf_float(result.classification[ix].value);
			  ei_printf("\n\n\n");
		  }
	  }
  }

I don’t really know what is missing and my model output looks like that:

Hilfe_EdgeImpulse
Hilfe_EdgeImpulse1041×540 32.2 KB

In case you’re wondering I’m french so robinet means faucet :slight_smile:. I’d appreciate any help or any clue. Thanks in advance.

Rene

#2

Hi @ReneW,

I’m not familiar with how BSP_AUDIO_IN_Init() works, but I might recommend creating a simple application that records a 1-2 sec clip of audio, print out the raw values, then save them as a wav file (e.g. copy/paste them into a short Python program that converts an array to a wav file: Python Convert an array to Wav - Stack Overflow). Play the .wav file to see if it sounds like what you intended.

I don’t know if it helps, but I have an example here that uses I2S with DMA to do continuous audio classification: ei-keyword-spotting/embedded-demos/stm32cubeide/nucleo-l476-keyword-spotting/Core/Src/main.cpp at a738973eb63f0818f2e0d420b05bfffe5dde06bb · ShawnHymel/ei-keyword-spotting · GitHub

#3

Hi,

sorry for the late response, I was sick. Thanks for the help @shawn_edgeimpulse. So I tried what you suggested and with the outputs of my values I can’t make a .wav-file (it’s created but can’t be played, I guess because it can’t work with the values). I think i have solved the issue with the DMA. When opening the .ioc-file I able to create and connect the DMA settings to the DFSDM1. The issue I’m facing right now though is that I’m not entering my Callback-functions anymore (both). I don’t know why it doens’t work anymore. Does maybe anyone have a clue?

I will try with the I2S in the meantime, to see if I can make it work.

Rene