diff --git a/tensorflow/lite/micro/examples/micro_speech/apollo3evb/audio_provider.cc b/tensorflow/lite/micro/examples/micro_speech/apollo3evb/audio_provider.cc
index 3cabf724b22..7d6d268bcc6 100644
--- a/tensorflow/lite/micro/examples/micro_speech/apollo3evb/audio_provider.cc
+++ b/tensorflow/lite/micro/examples/micro_speech/apollo3evb/audio_provider.cc
@@ -251,7 +251,6 @@ void pdm_start_dma(tflite::ErrorReporter* error_reporter) {
 
   // Reset the PDM DMA flags.
   g_pdm_dma_error = false;
-  g_pdm_dma_error_reporter = error_reporter;
 }
 
 #if USE_MAYA
@@ -470,11 +469,12 @@ TfLiteStatus InitAudioRecording(tflite::ErrorReporter* error_reporter) {
 #endif  // USE_TIME_STAMP
 
   // Configure, turn on PDM
+  g_pdm_dma_error_reporter = error_reporter;
   pdm_init();
   am_hal_interrupt_master_enable();
   am_hal_pdm_fifo_flush(g_pdm_handle);
   // Trigger the PDM DMA for the first time manually.
-  pdm_start_dma(g_pdm_dma_error_reporter);
+  pdm_start_dma(error_reporter);
 
   TF_LITE_REPORT_ERROR(error_reporter, "\nPDM DMA Threshold = %d",
                        PDMn(0)->FIFOTHR);