esphome 2024.6.6__py3-none-any.whl → 2024.7.0__py3-none-any.whl

esphome/components/micro_wake_word/micro_wake_word.h

@@ -1,21 +1,18 @@
 #pragma once
 
-/**
- * This is a workaround until we can figure out a way to get
- * the tflite-micro idf component code available in CI
- *
- * */
-//
-#ifndef CLANG_TIDY
-
 #ifdef USE_ESP_IDF
 
+#include "preprocessor_settings.h"
+#include "streaming_model.h"
+
 #include "esphome/core/automation.h"
 #include "esphome/core/component.h"
 #include "esphome/core/ring_buffer.h"
 
 #include "esphome/components/microphone/microphone.h"
 
+#include <frontend_util.h>
+
 #include <tensorflow/lite/core/c/common.h>
 #include <tensorflow/lite/micro/micro_interpreter.h>
 #include <tensorflow/lite/micro/micro_mutable_op_resolver.h>
@@ -23,35 +20,6 @@
 namespace esphome {
 namespace micro_wake_word {
 
-// The following are dictated by the preprocessor model
-//
-// The number of features the audio preprocessor generates per slice
-static const uint8_t PREPROCESSOR_FEATURE_SIZE = 40;
-// How frequently the preprocessor generates a new set of features
-static const uint8_t FEATURE_STRIDE_MS = 20;
-// Duration of each slice used as input into the preprocessor
-static const uint8_t FEATURE_DURATION_MS = 30;
-// Audio sample frequency in hertz
-static const uint16_t AUDIO_SAMPLE_FREQUENCY = 16000;
-// The number of old audio samples that are saved to be part of the next feature window
-static const uint16_t HISTORY_SAMPLES_TO_KEEP =
-    ((FEATURE_DURATION_MS - FEATURE_STRIDE_MS) * (AUDIO_SAMPLE_FREQUENCY / 1000));
-// The number of new audio samples to receive to be included with the next feature window
-static const uint16_t NEW_SAMPLES_TO_GET = (FEATURE_STRIDE_MS * (AUDIO_SAMPLE_FREQUENCY / 1000));
-// The total number of audio samples included in the feature window
-static const uint16_t SAMPLE_DURATION_COUNT = FEATURE_DURATION_MS * AUDIO_SAMPLE_FREQUENCY / 1000;
-// Number of bytes in memory needed for the preprocessor arena
-static const uint32_t PREPROCESSOR_ARENA_SIZE = 9528;
-
-// The following configure the streaming wake word model
-//
-// The number of audio slices to process before accepting a positive detection
-static const uint8_t MIN_SLICES_BEFORE_DETECTION = 74;
-
-// Number of bytes in memory needed for the streaming wake word model
-static const uint32_t STREAMING_MODEL_ARENA_SIZE = 64000;
-static const uint32_t STREAMING_MODEL_VARIABLE_ARENA_SIZE = 1024;
-
 enum State {
   IDLE,
   START_MICROPHONE,
@@ -61,6 +29,9 @@ enum State {
   STOPPING_MICROPHONE,
 };
 
+// The number of audio slices to process before accepting a positive detection
+static const uint8_t MIN_SLICES_BEFORE_DETECTION = 74;
+
 class MicroWakeWord : public Component {
  public:
   void setup() override;
@@ -73,28 +44,21 @@ class MicroWakeWord : public Component {
 
   bool is_running() const { return this->state_ != State::IDLE; }
 
-  bool initialize_models();
-
-  std::string get_wake_word() { return this->wake_word_; }
-
-  // Increasing either of these will reduce the rate of false acceptances while increasing the false rejection rate
-  void set_probability_cutoff(float probability_cutoff) { this->probability_cutoff_ = probability_cutoff; }
-  void set_sliding_window_average_size(size_t size);
+  void set_features_step_size(uint8_t step_size) { this->features_step_size_ = step_size; }
 
   void set_microphone(microphone::Microphone *microphone) { this->microphone_ = microphone; }
 
   Trigger<std::string> *get_wake_word_detected_trigger() const { return this->wake_word_detected_trigger_; }
 
-  void set_model_start(const uint8_t *model_start) { this->model_start_ = model_start; }
-  void set_wake_word(const std::string &wake_word) { this->wake_word_ = wake_word; }
-
- protected:
-  void set_state_(State state);
-  int read_microphone_();
+  void add_wake_word_model(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_average_size,
+                           const std::string &wake_word, size_t tensor_arena_size);
 
-  const uint8_t *model_start_;
-  std::string wake_word_;
+#ifdef USE_MICRO_WAKE_WORD_VAD
+  void add_vad_model(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_size,
+                     size_t tensor_arena_size);
+#endif
 
+ protected:
   microphone::Microphone *microphone_{nullptr};
   Trigger<std::string> *wake_word_detected_trigger_ = new Trigger<std::string>();
   State state_{State::IDLE};
@@ -102,85 +66,93 @@ class MicroWakeWord : public Component {
 
   std::unique_ptr<RingBuffer> ring_buffer_;
 
-  int16_t *input_buffer_;
+  std::vector<WakeWordModel> wake_word_models_;
 
-  const tflite::Model *preprocessor_model_{nullptr};
-  const tflite::Model *streaming_model_{nullptr};
-  tflite::MicroInterpreter *streaming_interpreter_{nullptr};
-  tflite::MicroInterpreter *preprocessor_interperter_{nullptr};
+#ifdef USE_MICRO_WAKE_WORD_VAD
+  std::unique_ptr<VADModel> vad_model_;
+#endif
 
-  std::vector<float> recent_streaming_probabilities_;
-  size_t last_n_index_{0};
+  tflite::MicroMutableOpResolver<20> streaming_op_resolver_;
 
-  float probability_cutoff_{0.5};
-  size_t sliding_window_average_size_{10};
+  // Audio frontend handles generating spectrogram features
+  struct FrontendConfig frontend_config_;
+  struct FrontendState frontend_state_;
 
-  // When the wake word detection first starts or after the word has been detected once, we ignore this many audio
-  // feature slices before accepting a positive detection again
+  // When the wake word detection first starts, we ignore this many audio
+  // feature slices before accepting a positive detection
   int16_t ignore_windows_{-MIN_SLICES_BEFORE_DETECTION};
 
-  uint8_t *streaming_var_arena_{nullptr};
-  uint8_t *streaming_tensor_arena_{nullptr};
-  uint8_t *preprocessor_tensor_arena_{nullptr};
-  int8_t *new_features_data_{nullptr};
-
-  tflite::MicroResourceVariables *mrv_{nullptr};
+  uint8_t features_step_size_;
 
-  // Stores audio fed into feature generator preprocessor
-  int16_t *preprocessor_audio_buffer_;
+  // Stores audio read from the microphone before being added to the ring buffer.
+  int16_t *input_buffer_{nullptr};
+  // Stores audio to be fed into the audio frontend for generating features.
+  int16_t *preprocessor_audio_buffer_{nullptr};
 
   bool detected_{false};
+  std::string detected_wake_word_{""};
 
-  /** Detects if wake word has been said
-   *
-   * If enough audio samples are available, it will generate one slice of new features.
-   * If the streaming model predicts the wake word, then the nonstreaming model confirms it.
-   * @param ring_Buffer Ring buffer containing raw audio samples
-   * @return True if the wake word is detected, false otherwise
-   */
-  bool detect_wake_word_();
+  void set_state_(State state);
 
-  /// @brief Returns true if there are enough audio samples in the buffer to generate another slice of features
-  bool slice_available_();
+  /// @brief Tests if there are enough samples in the ring buffer to generate new features.
+  /// @return True if enough samples, false otherwise.
+  bool has_enough_samples_();
 
-  /** Shifts previous feature slices over by one and generates a new slice of features
+  /** Reads audio from microphone into the ring buffer
    *
-   * @param ring_buffer ring buffer containing raw audio samples
-   * @return True if a new slice of features was generated, false otherwise
+   * Audio data (16000 kHz with int16 samples) is read into the input_buffer_.
+   * Verifies the ring buffer has enough space for all audio data. If not, it logs
+   * a warning and resets the ring buffer entirely.
+   * @return Number of bytes written to the ring buffer
    */
-  bool update_features_();
+  size_t read_microphone_();
+
+  /// @brief Allocates memory for input_buffer_, preprocessor_audio_buffer_, and ring_buffer_
+  /// @return True if successful, false otherwise
+  bool allocate_buffers_();
+
+  /// @brief Frees memory allocated for input_buffer_ and preprocessor_audio_buffer_
+  void deallocate_buffers_();
 
-  /** Generates features from audio samples
+  /// @brief Loads streaming models and prepares the feature generation frontend
+  /// @return True if successful, false otherwise
+  bool load_models_();
+
+  /// @brief Deletes each model's TFLite interpreters and frees tensor arena memory. Frees memory used by the feature
+  /// generation frontend.
+  void unload_models_();
+
+  /** Performs inference with each configured model
    *
-   * Adapted from TFLite micro speech example
-   * @param audio_data Pointer to array with the audio samples
-   * @param audio_data_size The number of samples to use as input to the preprocessor model
-   * @param feature_output Array that will store the features
-   * @return True if successful, false otherwise.
+   * If enough audio samples are available, it will generate one slice of new features.
+   * It then loops through and performs inference with each of the loaded models.
    */
-  bool generate_single_feature_(const int16_t *audio_data, int audio_data_size,
-                                int8_t feature_output[PREPROCESSOR_FEATURE_SIZE]);
+  void update_model_probabilities_();
 
-  /** Performs inference over the most recent feature slice with the streaming model
+  /** Checks every model's recent probabilities to determine if the wake word has been predicted
    *
-   * @return Probability of the wake word between 0.0 and 1.0
+   * Verifies the models have processed enough new samples for accurate predictions.
+   * Sets detected_wake_word_ to the wake word, if one is detected.
+   * @return True if a wake word is predicted, false otherwise
    */
-  float perform_streaming_inference_();
+  bool detect_wake_words_();
 
-  /** Strides the audio samples by keeping the last 10 ms of the previous slice
+  /** Generates features for a window of audio samples
    *
-   * Adapted from the TFLite micro speech example
-   * @param ring_buffer Ring buffer containing raw audio samples
-   * @param audio_samples Pointer to an array that will store the strided audio samples
-   * @return True if successful, false otherwise
+   * Reads samples from the ring buffer and feeds them into the preprocessor frontend.
+   * Adapted from TFLite microspeech frontend.
+   * @param features int8_t array to store the audio features
+   * @return True if successful, false otherwise.
    */
-  bool stride_audio_samples_(int16_t **audio_samples);
+  bool generate_features_for_window_(int8_t features[PREPROCESSOR_FEATURE_SIZE]);
 
-  /// @brief Returns true if successfully registered the preprocessor's TensorFlow operations
-  bool register_preprocessor_ops_(tflite::MicroMutableOpResolver<18> &op_resolver);
+  /// @brief Resets the ring buffer, ignore_windows_, and sliding window probabilities
+  void reset_states_();
 
   /// @brief Returns true if successfully registered the streaming model's TensorFlow operations
-  bool register_streaming_ops_(tflite::MicroMutableOpResolver<17> &op_resolver);
+  bool register_streaming_ops_(tflite::MicroMutableOpResolver<20> &op_resolver);
+
+  inline uint16_t new_samples_to_get_() { return (this->features_step_size_ * (AUDIO_SAMPLE_FREQUENCY / 1000)); }
 };
 
 template<typename... Ts> class StartAction : public Action<Ts...>, public Parented<MicroWakeWord> {
@@ -202,5 +174,3 @@ template<typename... Ts> class IsRunningCondition : public Condition<Ts...>, pub
 }  // namespace esphome
 
 #endif  // USE_ESP_IDF
-
-#endif  // CLANG_TIDY

esphome/components/micro_wake_word/preprocessor_settings.h

@@ -0,0 +1,20 @@
+#pragma once
+
+#ifdef USE_ESP_IDF
+
+#include <cstdint>
+
+namespace esphome {
+namespace micro_wake_word {
+
+// The number of features the audio preprocessor generates per slice
+static const uint8_t PREPROCESSOR_FEATURE_SIZE = 40;
+// Duration of each slice used as input into the preprocessor
+static const uint8_t FEATURE_DURATION_MS = 30;
+// Audio sample frequency in hertz
+static const uint16_t AUDIO_SAMPLE_FREQUENCY = 16000;
+
+}  // namespace micro_wake_word
+}  // namespace esphome
+
+#endif

esphome/components/micro_wake_word/streaming_model.cpp

@@ -0,0 +1,189 @@
+#ifdef USE_ESP_IDF
+
+#include "streaming_model.h"
+
+#include "esphome/core/hal.h"
+#include "esphome/core/helpers.h"
+#include "esphome/core/log.h"
+
+static const char *const TAG = "micro_wake_word";
+
+namespace esphome {
+namespace micro_wake_word {
+
+void WakeWordModel::log_model_config() {
+  ESP_LOGCONFIG(TAG, "    - Wake Word: %s", this->wake_word_.c_str());
+  ESP_LOGCONFIG(TAG, "      Probability cutoff: %.3f", this->probability_cutoff_);
+  ESP_LOGCONFIG(TAG, "      Sliding window size: %d", this->sliding_window_size_);
+}
+
+void VADModel::log_model_config() {
+  ESP_LOGCONFIG(TAG, "    - VAD Model");
+  ESP_LOGCONFIG(TAG, "      Probability cutoff: %.3f", this->probability_cutoff_);
+  ESP_LOGCONFIG(TAG, "      Sliding window size: %d", this->sliding_window_size_);
+}
+
+bool StreamingModel::load_model(tflite::MicroMutableOpResolver<20> &op_resolver) {
+  ExternalRAMAllocator<uint8_t> arena_allocator(ExternalRAMAllocator<uint8_t>::ALLOW_FAILURE);
+
+  if (this->tensor_arena_ == nullptr) {
+    this->tensor_arena_ = arena_allocator.allocate(this->tensor_arena_size_);
+    if (this->tensor_arena_ == nullptr) {
+      ESP_LOGE(TAG, "Could not allocate the streaming model's tensor arena.");
+      return false;
+    }
+  }
+
+  if (this->var_arena_ == nullptr) {
+    this->var_arena_ = arena_allocator.allocate(STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+    if (this->var_arena_ == nullptr) {
+      ESP_LOGE(TAG, "Could not allocate the streaming model's variable tensor arena.");
+      return false;
+    }
+    this->ma_ = tflite::MicroAllocator::Create(this->var_arena_, STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+    this->mrv_ = tflite::MicroResourceVariables::Create(this->ma_, 20);
+  }
+
+  const tflite::Model *model = tflite::GetModel(this->model_start_);
+  if (model->version() != TFLITE_SCHEMA_VERSION) {
+    ESP_LOGE(TAG, "Streaming model's schema is not supported");
+    return false;
+  }
+
+  if (this->interpreter_ == nullptr) {
+    this->interpreter_ = make_unique<tflite::MicroInterpreter>(
+        tflite::GetModel(this->model_start_), op_resolver, this->tensor_arena_, this->tensor_arena_size_, this->mrv_);
+    if (this->interpreter_->AllocateTensors() != kTfLiteOk) {
+      ESP_LOGE(TAG, "Failed to allocate tensors for the streaming model");
+      return false;
+    }
+
+    // Verify input tensor matches expected values
+    // Dimension 3 will represent the first layer stride, so skip it may vary
+    TfLiteTensor *input = this->interpreter_->input(0);
+    if ((input->dims->size != 3) || (input->dims->data[0] != 1) ||
+        (input->dims->data[2] != PREPROCESSOR_FEATURE_SIZE)) {
+      ESP_LOGE(TAG, "Streaming model tensor input dimensions has improper dimensions.");
+      return false;
+    }
+
+    if (input->type != kTfLiteInt8) {
+      ESP_LOGE(TAG, "Streaming model tensor input is not int8.");
+      return false;
+    }
+
+    // Verify output tensor matches expected values
+    TfLiteTensor *output = this->interpreter_->output(0);
+    if ((output->dims->size != 2) || (output->dims->data[0] != 1) || (output->dims->data[1] != 1)) {
+      ESP_LOGE(TAG, "Streaming model tensor output dimension is not 1x1.");
+    }
+
+    if (output->type != kTfLiteUInt8) {
+      ESP_LOGE(TAG, "Streaming model tensor output is not uint8.");
+      return false;
+    }
+  }
+
+  return true;
+}
+
+void StreamingModel::unload_model() {
+  this->interpreter_.reset();
+
+  ExternalRAMAllocator<uint8_t> arena_allocator(ExternalRAMAllocator<uint8_t>::ALLOW_FAILURE);
+
+  arena_allocator.deallocate(this->tensor_arena_, this->tensor_arena_size_);
+  this->tensor_arena_ = nullptr;
+  arena_allocator.deallocate(this->var_arena_, STREAMING_MODEL_VARIABLE_ARENA_SIZE);
+  this->var_arena_ = nullptr;
+}
+
+bool StreamingModel::perform_streaming_inference(const int8_t features[PREPROCESSOR_FEATURE_SIZE]) {
+  if (this->interpreter_ != nullptr) {
+    TfLiteTensor *input = this->interpreter_->input(0);
+
+    std::memmove(
+        (int8_t *) (tflite::GetTensorData<int8_t>(input)) + PREPROCESSOR_FEATURE_SIZE * this->current_stride_step_,
+        features, PREPROCESSOR_FEATURE_SIZE);
+    ++this->current_stride_step_;
+
+    uint8_t stride = this->interpreter_->input(0)->dims->data[1];
+
+    if (this->current_stride_step_ >= stride) {
+      this->current_stride_step_ = 0;
+
+      TfLiteStatus invoke_status = this->interpreter_->Invoke();
+      if (invoke_status != kTfLiteOk) {
+        ESP_LOGW(TAG, "Streaming interpreter invoke failed");
+        return false;
+      }
+
+      TfLiteTensor *output = this->interpreter_->output(0);
+
+      ++this->last_n_index_;
+      if (this->last_n_index_ == this->sliding_window_size_)
+        this->last_n_index_ = 0;
+      this->recent_streaming_probabilities_[this->last_n_index_] = output->data.uint8[0];  // probability;
+    }
+    return true;
+  }
+  ESP_LOGE(TAG, "Streaming interpreter is not initialized.");
+  return false;
+}
+
+void StreamingModel::reset_probabilities() {
+  for (auto &prob : this->recent_streaming_probabilities_) {
+    prob = 0;
+  }
+}
+
+WakeWordModel::WakeWordModel(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_average_size,
+                             const std::string &wake_word, size_t tensor_arena_size) {
+  this->model_start_ = model_start;
+  this->probability_cutoff_ = probability_cutoff;
+  this->sliding_window_size_ = sliding_window_average_size;
+  this->recent_streaming_probabilities_.resize(sliding_window_average_size, 0);
+  this->wake_word_ = wake_word;
+  this->tensor_arena_size_ = tensor_arena_size;
+};
+
+bool WakeWordModel::determine_detected() {
+  int32_t sum = 0;
+  for (auto &prob : this->recent_streaming_probabilities_) {
+    sum += prob;
+  }
+
+  float sliding_window_average = static_cast<float>(sum) / static_cast<float>(255 * this->sliding_window_size_);
+
+  // Detect the wake word if the sliding window average is above the cutoff
+  if (sliding_window_average > this->probability_cutoff_) {
+    ESP_LOGD(TAG, "The '%s' model sliding average probability is %.3f and most recent probability is %.3f",
+             this->wake_word_.c_str(), sliding_window_average,
+             this->recent_streaming_probabilities_[this->last_n_index_] / (255.0));
+    return true;
+  }
+  return false;
+}
+
+VADModel::VADModel(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_size,
+                   size_t tensor_arena_size) {
+  this->model_start_ = model_start;
+  this->probability_cutoff_ = probability_cutoff;
+  this->sliding_window_size_ = sliding_window_size;
+  this->recent_streaming_probabilities_.resize(sliding_window_size, 0);
+  this->tensor_arena_size_ = tensor_arena_size;
+};
+
+bool VADModel::determine_detected() {
+  uint8_t max = 0;
+  for (auto &prob : this->recent_streaming_probabilities_) {
+    max = std::max(prob, max);
+  }
+
+  return max > this->probability_cutoff_;
+}
+
+}  // namespace micro_wake_word
+}  // namespace esphome
+
+#endif

esphome/components/micro_wake_word/streaming_model.h

@@ -0,0 +1,84 @@
+#pragma once
+
+#ifdef USE_ESP_IDF
+
+#include "preprocessor_settings.h"
+
+#include <tensorflow/lite/core/c/common.h>
+#include <tensorflow/lite/micro/micro_interpreter.h>
+#include <tensorflow/lite/micro/micro_mutable_op_resolver.h>
+
+namespace esphome {
+namespace micro_wake_word {
+
+static const uint32_t STREAMING_MODEL_VARIABLE_ARENA_SIZE = 1024;
+
+class StreamingModel {
+ public:
+  virtual void log_model_config() = 0;
+  virtual bool determine_detected() = 0;
+
+  bool perform_streaming_inference(const int8_t features[PREPROCESSOR_FEATURE_SIZE]);
+
+  /// @brief Sets all recent_streaming_probabilities to 0
+  void reset_probabilities();
+
+  /// @brief Allocates tensor and variable arenas and sets up the model interpreter
+  /// @param op_resolver MicroMutableOpResolver object that must exist until the model is unloaded
+  /// @return True if successful, false otherwise
+  bool load_model(tflite::MicroMutableOpResolver<20> &op_resolver);
+
+  /// @brief Destroys the TFLite interpreter and frees the tensor and variable arenas' memory
+  void unload_model();
+
+ protected:
+  uint8_t current_stride_step_{0};
+
+  float probability_cutoff_;
+  size_t sliding_window_size_;
+  size_t last_n_index_{0};
+  size_t tensor_arena_size_;
+  std::vector<uint8_t> recent_streaming_probabilities_;
+
+  const uint8_t *model_start_;
+  uint8_t *tensor_arena_{nullptr};
+  uint8_t *var_arena_{nullptr};
+  std::unique_ptr<tflite::MicroInterpreter> interpreter_;
+  tflite::MicroResourceVariables *mrv_{nullptr};
+  tflite::MicroAllocator *ma_{nullptr};
+};
+
+class WakeWordModel final : public StreamingModel {
+ public:
+  WakeWordModel(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_average_size,
+                const std::string &wake_word, size_t tensor_arena_size);
+
+  void log_model_config() override;
+
+  /// @brief Checks for the wake word by comparing the mean probability in the sliding window with the probability
+  /// cutoff
+  /// @return True if wake word is detected, false otherwise
+  bool determine_detected() override;
+
+  const std::string &get_wake_word() const { return this->wake_word_; }
+
+ protected:
+  std::string wake_word_;
+};
+
+class VADModel final : public StreamingModel {
+ public:
+  VADModel(const uint8_t *model_start, float probability_cutoff, size_t sliding_window_size, size_t tensor_arena_size);
+
+  void log_model_config() override;
+
+  /// @brief Checks for voice activity by comparing the max probability in the sliding window with the probability
+  /// cutoff
+  /// @return True if voice activity is detected, false otherwise
+  bool determine_detected() override;
+};
+
+}  // namespace micro_wake_word
+}  // namespace esphome
+
+#endif

esphome/components/mitsubishi/mitsubishi.cpp

@@ -52,6 +52,7 @@ const uint8_t MITSUBISHI_BYTE16 = 0X00;
 
 climate::ClimateTraits MitsubishiClimate::traits() {
   auto traits = climate::ClimateTraits();
+  traits.set_supports_current_temperature(this->sensor_ != nullptr);
   traits.set_supports_action(false);
   traits.set_visual_min_temperature(MITSUBISHI_TEMP_MIN);
   traits.set_visual_max_temperature(MITSUBISHI_TEMP_MAX);

esphome/components/modbus_controller/text_sensor/__init__.py

@@ -37,6 +37,7 @@ RAW_ENCODING = {
     "NONE": RawEncoding.NONE,
     "HEXBYTES": RawEncoding.HEXBYTES,
     "COMMA": RawEncoding.COMMA,
+    "ANSI": RawEncoding.ANSI,
 }
 
 CONFIG_SCHEMA = cv.All(
@@ -49,7 +50,7 @@ CONFIG_SCHEMA = cv.All(
             cv.Optional(CONF_REGISTER_TYPE): cv.enum(MODBUS_REGISTER_TYPE),
             cv.Optional(CONF_REGISTER_COUNT, default=0): cv.positive_int,
             cv.Optional(CONF_RESPONSE_SIZE, default=2): cv.positive_int,
-            cv.Optional(CONF_RAW_ENCODE, default="NONE"): cv.enum(RAW_ENCODING),
+            cv.Optional(CONF_RAW_ENCODE, default="ANSI"): cv.enum(RAW_ENCODING),
         }
     ),
     validate_modbus_register,

esphome/components/modbus_controller/text_sensor/modbus_textsensor.cpp

@@ -27,8 +27,11 @@ void ModbusTextSensor::parse_and_publish(const std::vector<uint8_t> &data) {
         sprintf(buffer, index != this->offset ? ",%d" : "%d", b);
         output << buffer;
         break;
+      case RawEncoding::ANSI:
+        if (b < 0x20)
+          break;
+      // FALLTHROUGH
       // Anything else no encoding
-      case RawEncoding::NONE:
       default:
         output << (char) b;
         break;

esphome/components/modbus_controller/text_sensor/modbus_textsensor.h

@@ -9,7 +9,7 @@
 namespace esphome {
 namespace modbus_controller {
 
-enum class RawEncoding { NONE = 0, HEXBYTES = 1, COMMA = 2 };
+enum class RawEncoding { NONE = 0, HEXBYTES = 1, COMMA = 2, ANSI = 3 };
 
 class ModbusTextSensor : public Component, public text_sensor::TextSensor, public SensorItem {
  public:

esphome/components/number/__init__.py

@@ -26,6 +26,7 @@ from esphome.const import (
     DEVICE_CLASS_BATTERY,
     DEVICE_CLASS_CARBON_DIOXIDE,
     DEVICE_CLASS_CARBON_MONOXIDE,
+    DEVICE_CLASS_CONDUCTIVITY,
     DEVICE_CLASS_CURRENT,
     DEVICE_CLASS_DATA_RATE,
     DEVICE_CLASS_DATA_SIZE,
@@ -82,6 +83,7 @@ DEVICE_CLASSES = [
     DEVICE_CLASS_BATTERY,
     DEVICE_CLASS_CARBON_DIOXIDE,
     DEVICE_CLASS_CARBON_MONOXIDE,
+    DEVICE_CLASS_CONDUCTIVITY,
     DEVICE_CLASS_CURRENT,
     DEVICE_CLASS_DATA_RATE,
     DEVICE_CLASS_DATA_SIZE,