diff --git a/examples/device/uac2_speaker_fb/src/main.c b/examples/device/uac2_speaker_fb/src/main.c
index f2cfb155e..7b4e2d64c 100644
--- a/examples/device/uac2_speaker_fb/src/main.c
+++ b/examples/device/uac2_speaker_fb/src/main.c
@@ -35,13 +35,6 @@
 // MACRO CONSTANT TYPEDEF PROTOTYPES
 //--------------------------------------------------------------------+
 
-// List of supported sample rates for UAC2
-const uint32_t sample_rates[] = {44100, 48000, 88200, 96000};
-
-#define N_SAMPLE_RATES TU_ARRAY_SIZE(sample_rates)
-
-uint32_t current_sample_rate = 44100;
-
 /* Blink pattern
  * - 25 ms   : streaming data
  * - 250 ms  : device not mounted
@@ -76,6 +69,7 @@ static uint32_t blink_interval_ms = BLINK_NOT_MOUNTED;
 // Current states
 uint8_t mute[CFG_TUD_AUDIO_FUNC_1_N_CHANNELS_RX + 1];   // +1 for master channel 0
 int16_t volume[CFG_TUD_AUDIO_FUNC_1_N_CHANNELS_RX + 1];// +1 for master channel 0
+uint32_t current_sample_rate = 44100;
 
 // Buffer for speaker data
 uint16_t i2s_dummy_buffer[CFG_TUD_AUDIO_FUNC_1_EP_OUT_SW_BUF_SZ / 2];
@@ -316,6 +310,10 @@ static bool audio10_get_req_entity(uint8_t rhport, tusb_control_request_t const
 //--------------------------------------------------------------------+
 
 #if TUD_OPT_HIGH_SPEED
+// List of supported sample rates for UAC2
+const uint32_t sample_rates[] = {44100, 48000, 88200, 96000};
+
+#define N_SAMPLE_RATES TU_ARRAY_SIZE(sample_rates)
 
 static bool audio20_clock_get_request(uint8_t rhport, audio20_control_request_t const *request) {
   TU_ASSERT(request->bEntityID == UAC2_ENTITY_CLOCK);
@@ -548,6 +546,17 @@ void tud_audio_feedback_params_cb(uint8_t func_id, uint8_t alt_itf, audio_feedba
   // Set feedback method to fifo counting
   feedback_param->method = AUDIO_FEEDBACK_METHOD_FIFO_COUNT;
   feedback_param->sample_freq = current_sample_rate;
+
+  // About FIFO threshold:
+  //
+  // By default the threshold is set to half FIFO size, which works well in most cases,
+  // you can reduce the threshold to have less latency.
+  //
+  // For example, here we could set the threshold to 2 ms of audio data, as audio_task() read audio data every 1 ms,
+  // having 2 ms threshold allows some margin and a quick response:
+  //
+  // feedback_param->fifo_count.fifo_threshold =
+  //    current_sample_rate * CFG_TUD_AUDIO_FUNC_1_N_CHANNELS_RX * CFG_TUD_AUDIO_FUNC_1_N_BYTES_PER_SAMPLE_RX / 1000 * 2;
 }
 
 #if CFG_AUDIO_DEBUG
diff --git a/src/class/audio/audio_device.c b/src/class/audio/audio_device.c
index 328f085a8..32fc4d076 100644
--- a/src/class/audio/audio_device.c
+++ b/src/class/audio/audio_device.c
@@ -1546,7 +1546,7 @@ static bool audiod_fb_params_prepare(uint8_t func_id, uint8_t alt) {
 
   // Prepare feedback computation if endpoint is available
   if (audio->ep_fb != 0) {
-    audio_feedback_params_t fb_param;
+    audio_feedback_params_t fb_param = {0};
 
     tud_audio_feedback_params_cb(func_id, alt, &fb_param);
     audio->feedback.compute_method = fb_param.method;
@@ -1587,15 +1587,15 @@ static bool audiod_fb_params_prepare(uint8_t func_id, uint8_t alt) {
       } break;
 
       case AUDIO_FEEDBACK_METHOD_FIFO_COUNT: {
-        // Initialize the threshold level to half filled
-        uint16_t fifo_lvl_thr = tu_fifo_depth(&audio->ep_out_ff) / 2;
-        audio->feedback.compute.fifo_count.fifo_lvl_thr = fifo_lvl_thr;
-        audio->feedback.compute.fifo_count.fifo_lvl_avg = ((uint32_t) fifo_lvl_thr) << 16;
+        // Determine FIFO threshold
+        uint16_t fifo_threshold = fb_param.fifo_count.fifo_threshold ? fb_param.fifo_count.fifo_threshold : tu_fifo_depth(&audio->ep_out_ff) / 2;
+        audio->feedback.compute.fifo_count.fifo_lvl_thr = fifo_threshold;
+        audio->feedback.compute.fifo_count.fifo_lvl_avg = ((uint32_t) fifo_threshold) << 16;
         // Avoid 64bit division
         uint32_t nominal = ((fb_param.sample_freq / 100) << 16) / (frame_div / 100);
         audio->feedback.compute.fifo_count.nom_value = nominal;
-        audio->feedback.compute.fifo_count.rate_const[0] = (uint16_t) ((audio->feedback.max_value - nominal) / fifo_lvl_thr);
-        audio->feedback.compute.fifo_count.rate_const[1] = (uint16_t) ((nominal - audio->feedback.min_value) / fifo_lvl_thr);
+        audio->feedback.compute.fifo_count.rate_const[0] = (uint16_t) ((audio->feedback.max_value - nominal) / fifo_threshold);
+        audio->feedback.compute.fifo_count.rate_const[1] = (uint16_t) ((nominal - audio->feedback.min_value) / fifo_threshold);
         // On HS feedback is more sensitive since packet size can vary every MSOF, could cause instability
         if (tud_speed_get() == TUSB_SPEED_HIGH) {
           audio->feedback.compute.fifo_count.rate_const[0] /= 8;
diff --git a/src/class/audio/audio_device.h b/src/class/audio/audio_device.h
index ab3684aad..1e0c46915 100644
--- a/src/class/audio/audio_device.h
+++ b/src/class/audio/audio_device.h
@@ -333,10 +333,12 @@ typedef struct {
   union {
     struct {
       uint32_t mclk_freq; // Main clock frequency in Hz i.e. master clock to which sample clock is based on
-    }frequency;
-
+    } frequency;
+    struct {
+      uint16_t fifo_threshold;  // Target FIFO threshold level, default to half FIFO if not set
+    } fifo_count;
   };
-}audio_feedback_params_t;
+} audio_feedback_params_t;
 
 // Invoked when needed to set feedback parameters
 void tud_audio_feedback_params_cb(uint8_t func_id, uint8_t alt_itf, audio_feedback_params_t* feedback_param);