I have port the tflm example micro_speech(the latest code on git) on nordic 52840. All the codes work fine without any errors, but the invoke result is unexpected. the code as follow (port from micro_speech_test.cc):
int8_t expected_feature[kFeatureSize] = {
126, 103, 124, 102, 124, 102, 123, 100, 118, 97, 118, 100, 118, 98,
121, 100, 121, 98, 117, 91, 96, 74, 54, 87, 100, 87, 109, 92,
91, 80, 64, 55, 83, 74, 74, 78, 114, 95, 101, 81,
};
TF_LITE_ENSURE_STATUS(GenerateFeatures(
g_no_30ms_audio_data, g_no_30ms_audio_data_size, &g_features));
for (i = 0; i < kFeatureSize; i++) {
if(g_features[0][i] != expected_feature[i]) {
printk(“NoFeatureTest fail %d %d\n”, g_features[0][i], expected_feature[i]);
//return kTfLiteError;
}
=> I found that the expected_feature[0][i] are not same with tflm invoke result g_features[0][i]. what should i do ?