Initialization & Configuration

Configuration Structure

The SDK is configured through cry_detect_config_t (defined in cry_detect.h):

typedef struct {
    const char *engine_type;           // "pure_c" | "svp_acl" | "magik" | "int8"
    const char *weight_path;           // path to model weights (.bin / .om / .mgk)
    const char *capture_type;          // "mpp" | "imp" | "file"
    int         silence_rms_threshold; // 0 = disable, 50 = default
    cry_detect_callback_t  callback;   // result callback (can be NULL)
    void       *callback_data;         // user data passed to callback
    const char *alarm_strategy;        // "asymmetric_envelope" or NULL
    const char *alarm_params;          // "k1=v1,k2=v2" or NULL for defaults
    alarm_callback_t alarm_callback;   // alarm event callback
    void       *alarm_callback_data;   // alarm callback user data
} cry_detect_config_t;

All fields are optional — pass NULL (or 0) to use defaults. The default engine and capture type are determined at compile time by the build system (see config.h).

Initialization Sequence

// 1. Create config (or pass NULL for all defaults)
cry_detect_config_t config = {
    .engine_type  = "pure_c",
    .weight_path  = "./model_weights.bin",
    .callback     = my_callback,
    .callback_data = &my_context,
    .alarm_strategy = "asymmetric_envelope",
    .alarm_callback = my_alarm_callback,
};

// 2. Initialize
int ret = cry_detect_init(&config);
if (ret < 0) {
    // Handle error (e.g., model file not found, unsupported engine)
    return ret;
}

// 3. Start real-time detection
ret = cry_detect_start();
if (ret < 0) {
    // Handle error (e.g., audio device unavailable)
    cry_detect_deinit();
    return ret;
}

// 4. Detection runs in background — results arrive via callbacks
// Your application can do other work here

// 5. Stop detection
cry_detect_stop();

// 6. Clean up
cry_detect_deinit();

Compile-Time Configuration (`config.h`)

Audio parameters are defined in config.h and must match the training configuration:

Platform-specific audio parameters (device IDs, frame sizes) are selected automatically based on the AUDIO_CAPTURE_* macro.

Default Backend Selection

The build script defines which backends are compiled in:

File Processing Mode

For testing with WAV files (no microphone needed):

cry_detect_init(NULL);  // all defaults
cry_detect_process_file("test.wav");
cry_detect_deinit();

This is useful for offline testing on a PC or during CI.