AI - Feature Extract APIs

Introduction

This contains prototyping of user interactive API functions

Files
file	feature_extract.h
	Header file contains enumerations, structure definitions and functions declarations for Edge AI Feature Extract Module interface.

Data Structures
struct	feature_extraction
	Feature extraction context structure. More...

Functions
void	FE_allocFeatureExtract (feature_extraction_handle fe)
	Calculates required memory allocation for feature extraction.
void	FE_initFeatureExtract (feature_extraction_handle fe)
	Initializes feature extraction state and buffers.
void	FE_runFeatureExtract (feature_extraction_handle fe)
	Main feature extraction processing function.
int	FE_compareModelInput (model_input_t *expected_array, model_input_t *obtained_array)
	Compares feature extraction buffers.
int	FE_compareModelOutput (model_output_t *expected_array, model_output_t *obtained_array)
	Compares nn model buffers.

Typedefs
typedef float	model_input_t
	Model Input.
typedef float	model_output_t
	Model Output.
typedef feature_extraction *	feature_extraction_handle

Typedef Documentation

model_input_t

typedef float model_input_t

Model Input.

Model Input type

model_output_t

typedef float model_output_t

Model Output.

Model Output type

feature_extraction_handle

typedef feature_extraction* feature_extraction_handle

Function Documentation

FE_allocFeatureExtract()

void FE_allocFeatureExtract

(

feature_extraction_handle

fe

)

Calculates required memory allocation for feature extraction.

Parameters

fe	Feature extraction handle

Sets the required memory size for feature extraction operations Size is calculated as 2x frame size in float precision

Returns

None

FE_initFeatureExtract()

void FE_initFeatureExtract

(

feature_extraction_handle

fe

)

Initializes feature extraction state and buffers.

Parameters

fe	Feature extraction handle

Initializes history buffer and output buffer for feature extraction Handles both normalized and quantized output modes Processes each channel separately for multi-channel data

Returns

None

FE_runFeatureExtract()

void FE_runFeatureExtract

(

feature_extraction_handle

fe

)

Main feature extraction processing function.

Parameters

fe	Feature extraction handle containing all buffers and state

Performs complete feature extraction pipeline including:

• Raw signal processing (optional mean removal)
• Windowing (optional)
• FFT computation (optional)
• FFT normalization (optional)
• DC removal (optional)
• Frequency binning (optional)
• Log scaling (optional)
• Frame concatenation (optional)
• Batch normalization or direct copy (based on configuration)

Processes each channel independently and supports both normalized float output and quantized int8_t output modes

Returns

None

FE_compareModelInput()

int FE_compareModelInput	(	model_input_t *	expected_array,
		model_input_t *	obtained_array )

Compares feature extraction buffers.

Parameters

expected_array	Expected output from feature extraction
obtained_array	Obtained output from feature extraction

Takes the difference of the two arrays and returns the number of elements that were not matched.

Returns

number of mismatches in two array

FE_compareModelOutput()

int FE_compareModelOutput	(	model_output_t *	expected_array,
		model_output_t *	obtained_array )

Compares nn model buffers.

Parameters

expected_array	Expected output from nn model
obtained_array	Obtained output from nn model

Takes the difference of the two arrays and returns the number of elements that were not matched.

Returns

number of mismatches in two array