Microphone and mWW Improvements (#43)

* support 48kHz hacked decimation and secondary microphone stream

* use task notifications to manage mic

* don't use all of a core for mic reading

* cleanup and simplify logic

* add available function to secondary mic

* move mww into tasks

* use event group to manage tasks

* simplify component logic

* update some TODOs

* use quantized probability cutoff for comparison

* move ignore_windows into each model

* pass more information about detected wake word

* use a queue to send wake word info from task to loop

* enable/disable particular wake word models

* disable task debug logging

* add enabled wake word selector

* yamllint fix

* revert removing a mww start, you'll get a warning

* add short delay after stopping ble

esphome/components/i2s_audio/microphone/i2s_audio_microphone.h

@@ -14,9 +14,6 @@
 namespace esphome {
 namespace i2s_audio {
 
-
-
-
 class I2SAudioMicrophone : public I2SAudioIn, public microphone::Microphone, public Component {
  public:
   void setup() override;
@@ -25,10 +22,10 @@ class I2SAudioMicrophone : public I2SAudioIn, public microphone::Microphone, pub
 
   void loop() override;
 
-  void set_din_pin(int8_t pin) { this->din_pin_ = pin; }
-  void set_pdm(bool pdm) { this->pdm_ = pdm; }
-
   size_t read(int16_t *buf, size_t len) override;
+  size_t read_secondary(int16_t *buf, size_t len) override;
+
+  size_t available_secondary() override { return this->comm_ring_buffer_->available(); }
 
 #if SOC_I2S_SUPPORTS_ADC
   void set_adc_channel(adc1_channel_t channel) {
@@ -42,29 +39,31 @@ class I2SAudioMicrophone : public I2SAudioIn, public microphone::Microphone, pub
   void set_bits_per_sample(i2s_bits_per_sample_t bits_per_sample) { this->bits_per_sample_ = bits_per_sample; }
   void set_use_apll(uint32_t use_apll) { this->use_apll_ = use_apll; }
 
+  void set_din_pin(int8_t pin) { this->din_pin_ = pin; }
+  void set_pdm(bool pdm) { this->pdm_ = pdm; }
+
  protected:
+  esp_err_t start_i2s_driver_();
+
   static void read_task_(void *params);
-  void watch_();
 
   TaskHandle_t read_task_handle_{nullptr};
   QueueHandle_t event_queue_;
-  QueueHandle_t command_queue_;
-  std::unique_ptr<RingBuffer> output_ring_buffer_;
-
-  void start_();
-  void stop_();
-  void read_();
+  std::unique_ptr<RingBuffer> asr_ring_buffer_;
+  std::unique_ptr<RingBuffer> comm_ring_buffer_;
 
+  bool use_apll_;
+  bool pdm_{false};
   int8_t din_pin_{I2S_PIN_NO_CHANGE};
+
 #if SOC_I2S_SUPPORTS_ADC
-  adc1_channel_t adc_channel_{ADC1_CHANNEL_MAX};
   bool adc_{false};
+  adc1_channel_t adc_channel_{ADC1_CHANNEL_MAX};
 #endif
-  bool pdm_{false};
+
+  i2s_bits_per_sample_t bits_per_sample_;
   i2s_channel_fmt_t channel_;
   uint32_t sample_rate_;
-  i2s_bits_per_sample_t bits_per_sample_;
-  bool use_apll_;
 };
 
 }  // namespace i2s_audio

esphome/components/micro_wake_word/__init__.py

@@ -64,6 +64,8 @@
     "IsRunningCondition", automation.Condition
 )
 
+WakeWordModel_ = micro_wake_word_ns.class_("WakeWordModel")
+
 
 def _validate_json_filename(value):
     value = cv.string(value)
@@ -324,14 +326,15 @@ def _validate_source_shorthand(value):
 
 MODEL_SCHEMA = cv.Schema(
     {
+        cv.GenerateID(CONF_ID): cv.declare_id(WakeWordModel_),
         cv.Optional(CONF_MODEL): MODEL_SOURCE_SCHEMA,
         cv.Optional(CONF_PROBABILITY_CUTOFF): cv.percentage,
         cv.Optional(CONF_SLIDING_WINDOW_SIZE): cv.positive_int,
         cv.GenerateID(CONF_RAW_DATA_ID): cv.declare_id(cg.uint8),
     }
 )
 
-# Provide a default VAD model that could be overridden
+# Provides a default VAD model that could be overridden
 VAD_MODEL_SCHEMA = MODEL_SCHEMA.extend(
     cv.Schema(
         {
@@ -476,6 +479,8 @@ async def to_code(config):
         probability_cutoff = model_parameters.get(
             CONF_PROBABILITY_CUTOFF, manifest[KEY_MICRO][CONF_PROBABILITY_CUTOFF]
         )
+        quantized_probability_cutoff = int(probability_cutoff * 255)
+
         sliding_window_size = model_parameters.get(
             CONF_SLIDING_WINDOW_SIZE,
             manifest[KEY_MICRO][CONF_SLIDING_WINDOW_SIZE],
@@ -485,24 +490,29 @@ async def to_code(config):
             cg.add(
                 var.add_vad_model(
                     prog_arr,
-                    probability_cutoff,
+                    quantized_probability_cutoff,
                     sliding_window_size,
                     manifest[KEY_MICRO][CONF_TENSOR_ARENA_SIZE],
                 )
             )
         else:
-            cg.add(
-                var.add_wake_word_model(
-                    prog_arr,
-                    probability_cutoff,
-                    sliding_window_size,
-                    manifest[KEY_WAKE_WORD],
-                    manifest[KEY_MICRO][CONF_TENSOR_ARENA_SIZE],
-                )
+            wake_word_model = cg.new_Pvariable(
+                model_parameters[CONF_ID],
+                prog_arr,
+                quantized_probability_cutoff,
+                sliding_window_size,
+                manifest[KEY_WAKE_WORD],
+                manifest[KEY_MICRO][CONF_TENSOR_ARENA_SIZE],
             )
 
+            cg.add(var.add_wake_word_model(wake_word_model))
+
     cg.add(var.set_features_step_size(manifest[KEY_MICRO][CONF_FEATURE_STEP_SIZE]))
-    cg.add_library(None,None,"https://github.com/kahrendt/ESPMicroSpeechFeatures.git#psram-allocations")
+    cg.add_library(
+        None,
+        None,
+        "https://github.com/kahrendt/ESPMicroSpeechFeatures.git#psram-allocations",
+    )
     # cg.add_library("kahrendt/ESPMicroSpeechFeatures", "1.0.0")