diff --git a/esphome/components/speaker/__init__.py b/esphome/components/speaker/__init__.py
index 5f1ba94ee6..18e1d9782c 100644
--- a/esphome/components/speaker/__init__.py
+++ b/esphome/components/speaker/__init__.py
@@ -3,7 +3,7 @@ import esphome.codegen as cg
 from esphome.components import audio, audio_dac
 import esphome.config_validation as cv
 from esphome.const import CONF_DATA, CONF_ID, CONF_VOLUME
-from esphome.core import CORE
+from esphome.core import CORE, ID
 from esphome.coroutine import CoroPriority, coroutine_with_priority
 
 AUTO_LOAD = ["audio"]
@@ -90,7 +90,10 @@ async def speaker_play_action(config, action_id, template_arg, args):
         templ = await cg.templatable(data, args, cg.std_vector.template(cg.uint8))
         cg.add(var.set_data_template(templ))
     else:
-        cg.add(var.set_data_static(data))
+        # Generate static array in flash to avoid RAM copy
+        arr_id = ID(f"{action_id}_data", is_declaration=True, type=cg.uint8)
+        arr = cg.static_const_array(arr_id, cg.ArrayInitializer(*data))
+        cg.add(var.set_data_static(arr, len(data)))
     return var
 
 
diff --git a/esphome/components/speaker/automation.h b/esphome/components/speaker/automation.h
index 80bba25030..cae429cf8a 100644
--- a/esphome/components/speaker/automation.h
+++ b/esphome/components/speaker/automation.h
@@ -10,28 +10,35 @@ namespace speaker {
 
 template<typename... Ts> class PlayAction : public Action<Ts...>, public Parented<Speaker> {
  public:
-  void set_data_template(std::function<std::vector<uint8_t>(Ts...)> func) {
-    this->data_func_ = func;
+  void set_data_template(std::vector<uint8_t> (*func)(Ts...)) {
+    this->data_.func = func;
     this->static_ = false;
   }
-  void set_data_static(const std::vector<uint8_t> &data) {
-    this->data_static_ = data;
+
+  void set_data_static(const uint8_t *data, size_t len) {
+    this->data_.static_data.ptr = data;
+    this->data_.static_data.len = len;
     this->static_ = true;
   }
 
   void play(const Ts &...x) override {
     if (this->static_) {
-      this->parent_->play(this->data_static_);
+      this->parent_->play(this->data_.static_data.ptr, this->data_.static_data.len);
     } else {
-      auto val = this->data_func_(x...);
+      auto val = this->data_.func(x...);
       this->parent_->play(val);
     }
   }
 
  protected:
-  bool static_{false};
-  std::function<std::vector<uint8_t>(Ts...)> data_func_{};
-  std::vector<uint8_t> data_static_{};
+  bool static_{true};
+  union Data {
+    std::vector<uint8_t> (*func)(Ts...);
+    struct {
+      const uint8_t *ptr;
+      size_t len;
+    } static_data;
+  } data_;
 };
 
 template<typename... Ts> class VolumeSetAction : public Action<Ts...>, public Parented<Speaker> {
diff --git a/tests/components/speaker/common.yaml b/tests/components/speaker/common.yaml
index c04674ee29..fa54fa7e39 100644
--- a/tests/components/speaker/common.yaml
+++ b/tests/components/speaker/common.yaml
@@ -1,3 +1,12 @@
+number:
+  - platform: template
+    name: "Speaker Number"
+    id: my_number
+    optimistic: true
+    min_value: 0
+    max_value: 100
+    step: 1
+
 esphome:
   on_boot:
     then:
@@ -14,6 +23,15 @@ esphome:
             - speaker.finish:
       - speaker.stop:
 
+button:
+  - platform: template
+    name: "Speaker Button"
+    on_press:
+      then:
+        - speaker.play: [0x10, 0x20, 0x30, 0x40]
+        - speaker.play: !lambda |-
+            return {0x01, 0x02, (uint8_t)id(my_number).state};
+
 i2s_audio:
   i2s_lrclk_pin: ${i2s_bclk_pin}
   i2s_bclk_pin: ${i2s_lrclk_pin}