I Always have nan and -nan values on MFCC output

[IFaye.1]

Hello Everyone,

I try to extract MFccColumn from my microphones plug on stm32h7. But when I put my buffer output audio on my function for extracting the mfcc I always get nan or -nan on pOutColBuffer[NUM_MELS].

Does someone have idea? Thanks,

Here the code:

#include "feature_extraction.h"

#include "AWEPlatform.h"

#define SAMPLE_RATE 16000U /* Input signal sampling rate */

#define FFT_LEN   192U /* Number of FFT points. Must be greater or equal to FRAME_LEN */

#define NUM_FRAMES  16U /* Number of columns in spectrogram */

#define FRAME_LEN FFT_LEN /* Window length and then padded with zeros to match FFT_LEN. */

#define HOP_LEN   92U /* Number of overlapping samples between successive frames. */

#define NUM_MELS   16U /* Number of mel bands */

float32_t pInBuffer[FRAME_LEN];           /* 8.0 KB */

float32_t pOutColBuffer[NUM_MELS];          /* 0.5 KB */

float32_t pOutMelSpectrogram[NUM_MELS * NUM_FRAMES]; /* 7.0 KB */

float32_t pSpectrScratchBuffer[FFT_LEN];       /* 8.0 KB */

float32_t pWindowFuncBuffer[FFT_LEN];        /* 8.0 KB */

uint32_t pMelFilterStartIndices[NUM_MELS];     /* 0.5 KB */

uint32_t pMelFilterStopIndices[NUM_MELS];      /* 0.5 KB */

float32_t pMelFilterCoefs[2020];           /* 7.9 KB */ /* Size given by S_MelFilter.CoefficientsLength */

/* Allocate buffers and structures */

arm_rfft_fast_instance_f32 S_Rfft;     /* 24 B */

MelFilterTypeDef      S_MelFilter;   /* 48 B */

SpectrogramTypeDef     S_Spectr;    /* 28 B */

MelSpectrogramTypeDef   S_MelSpectr;   /* 8 B */

LogMelSpectrogramTypeDef  S_LogMelSpectr; /* 16 B */

DCT_InstanceTypeDef S_Dctinstance;

MfccTypeDef S_Mfcc;

/*

 * Python equivalent:

 * librosa.feature.melspectrogram(y=y, sr=16000, n_mels=128, hop_length=1024, center=False)

 */

void Preprocessing_Init(void)

{

 /* Init window function */

 if (Window_Init(pWindowFuncBuffer, FRAME_LEN, WINDOW_HANN) != 0)

 {

  printf("Init error\n");

  exit(1);

 }

 /* Init RFFT */

 arm_rfft_fast_init_f32(&S_Rfft, FFT_LEN);

 /* Init Mel filter */

 S_MelFilter.pStartIndices = (uint32_t *)pMelFilterStartIndices;

 S_MelFilter.pStopIndices = (uint32_t *)pMelFilterStopIndices;

 S_MelFilter.pCoefficients = (uint32_t *)pMelFilterCoefs;

 S_MelFilter.NumMels  = NUM_MELS;

 S_MelFilter.FFTLen  = FFT_LEN;

 S_MelFilter.SampRate = SAMPLE_RATE;

 S_MelFilter.FMin   = 0.0;

 S_MelFilter.FMax   = S_MelFilter.SampRate / 2.0;

 S_MelFilter.Formula  = MEL_SLANEY;

 S_MelFilter.Normalize = 1;

 S_MelFilter.Mel2F   = 1;

 MelFilterbank_Init(&S_MelFilter);

 /* Init Spectrogram */

 S_Spectr.pRfft  = &S_Rfft;

 S_Spectr.Type   = SPECTRUM_TYPE_POWER;

 S_Spectr.pWindow = (float32_t *)pWindowFuncBuffer;

 S_Spectr.SampRate = SAMPLE_RATE;

 S_Spectr.FrameLen = FRAME_LEN;

 S_Spectr.FFTLen  = FFT_LEN;

 S_Spectr.pScratch = pSpectrScratchBuffer;

 /* Init MelSpectrogram */

 S_MelSpectr.SpectrogramConf = &S_Spectr;

 S_MelSpectr.MelFilter    = &S_MelFilter;

 /* LogMelSpectr */

 S_LogMelSpectr.LogFormula = LOGMELSPECTROGRAM_SCALE_DB;

 S_LogMelSpectr.MelSpectrogramConf = &S_MelSpectr;

 S_LogMelSpectr.Ref = 1.0;

 S_LogMelSpectr.TopdB = -80;

 /* DCT instacne */

 float32_t pDCTCoefsBuffer[16 * 16];

 //float32_t pOutBuffer[13];

 S_Dctinstance.NumFilters = 16;

 S_Dctinstance.NumInputs = 16;

 S_Dctinstance.Type = DCT_TYPE_III;

 S_Dctinstance.RemoveDCTZero = 1;

 S_Dctinstance.pDCTCoefs = pDCTCoefsBuffer;

 DCT_Init(&S_Dctinstance);

 /* Init Mfcc */

 S_Mfcc.LogMelConf = &S_LogMelSpectr;

 S_Mfcc.pDCT = &S_Dctinstance;

 S_Mfcc.NumMfccCoefs = NUM_FRAMES;

 S_Mfcc.pScratch = pSpectrScratchBuffer;

}

void AudioPreprocessing_Run(int16_t *pInSignal)

{

buf_to_float(pInSignal,pInBuffer,FRAME_LEN);

MfccColumn(&S_Mfcc, pInBuffer, pOutColBuffer);

/* Reshape col into pOutMelSpectrogram */

  //for (uint32_t i = 0; i < NUM_MELS; i++)

//{

// pOutMelSpectrogram[i * NUM_FRAMES + frame_index] = pOutColBuffer[i];

//}

}

My code to get Microphones. Its works well:

void SetupAudio(void)

{

  memset(AudioBufferIn, 0, sizeof(AudioBufferIn));

  memset(AudioBufferOut, 0, sizeof(AudioBufferOut));

  BSP_DFSDM_IN_Init(INPUT_DEVICE_DIGITAL_MIC1, BSP_AUDIO_FREQUENCY_48K, MIC_SAMPLE_SIZE_IN_BITS, NUMBER_MICS);

  BSP_AUDIO_IN_OUT_Init(INPUT_DEVICE_INPUT_LINE_1, OUTPUT_DEVICE_HEADPHONE, BSP_AUDIO_FREQUENCY_48K);

}  // End SetupAudio

// Audio Init

void AudioInit(void)

{

// some code ...

  HAL_NVIC_SetPriority(AUDIO_IN_SAIx_DMAx_IRQ, 3, 0);

  HAL_NVIC_EnableIRQ(AUDIO_IN_SAIx_DMAx_IRQ);

  BSP_DFSDM_IN_Record(DFSDM_scratchBufferList, MIC_SINGLE_CHANNEL_BUFFER_SIZE);

  BSP_AUDIO_IN_Record( (UINT16 *)AudioBufferIn, INPUT_AUDIO_BUFFER_SIZE);

  BSP_AUDIO_OUT_Play( (UINT16 *)AudioBufferOut, OUTPUT_AUDIO_BUFFER_SIZE);

  //--- MFCC Init -----

  Preprocessing_Init();

} // End awe_pltInit

//-----------------------------------------------------------------------------

// METHOD: AUDIO_IN_DMA_Callback

// PURPOSE: Interrupt handler - called each time a frame of audio received

//-----------------------------------------------------------------------------

void AUDIO_IN_DMA_Callback(void)

{

// some codes ...

    //--- MFCC Compute -----

    AudioPreprocessing_Run(&AudioBufferOut[0]);

// some code ...

}  // End AUDIO_IN_DMA_Callback