voice_assistant.h

Go to the documentation of this file.

#pragma once
 
#include "esphome/core/defines.h"
 
#ifdef USE_VOICE_ASSISTANT
 
#include "esphome/core/automation.h"
#include "esphome/core/component.h"
#include "esphome/core/helpers.h"
 
#include "esphome/components/api/api_connection.h"
#include "esphome/components/audio/audio_transfer_buffer.h"
#include "esphome/components/ring_buffer/ring_buffer.h"
#include "esphome/components/api/api_pb2.h"
#include "esphome/components/microphone/microphone_source.h"
#ifdef USE_MEDIA_PLAYER
#include "esphome/components/media_player/media_player.h"
#endif
#ifdef USE_MICRO_WAKE_WORD
#include "esphome/components/micro_wake_word/micro_wake_word.h"
#endif
#ifdef USE_SPEAKER
#include "esphome/components/speaker/speaker.h"
#endif
#include "esphome/components/socket/socket.h"
 
#include <span>
#include <vector>
 
namespace esphome::voice_assistant {
 
// Version 1: Initial version
// Version 2: Adds raw speaker support
static const uint32_t LEGACY_INITIAL_VERSION = 1;
static const uint32_t LEGACY_SPEAKER_SUPPORT = 2;
 
enum VoiceAssistantFeature : uint32_t {
  FEATURE_VOICE_ASSISTANT = 1 << 0,
  FEATURE_SPEAKER = 1 << 1,
  FEATURE_API_AUDIO = 1 << 2,
  FEATURE_TIMERS = 1 << 3,
  FEATURE_ANNOUNCE = 1 << 4,
  FEATURE_START_CONVERSATION = 1 << 5,
  FEATURE_MULTI_CHANNEL_AUDIO = 1 << 6,
};
 
enum class State : uint8_t {
  IDLE,
  START_MICROPHONE,
  STARTING_MICROPHONE,
  WAIT_FOR_VAD,
  WAITING_FOR_VAD,
  START_PIPELINE,
  STARTING_PIPELINE,
  STREAMING_MICROPHONE,
  STOP_MICROPHONE,
  STOPPING_MICROPHONE,
  AWAITING_RESPONSE,
  STREAMING_RESPONSE,
  RESPONSE_FINISHED,
};
 
enum AudioMode : uint8_t {
  AUDIO_MODE_UDP,
  AUDIO_MODE_API,
};
 
struct Timer {
  std::string id;
  std::string name;
  uint32_t total_seconds;
  uint32_t seconds_left;
  bool is_active;
 
  static constexpr size_t TO_STR_BUFFER_SIZE = 128;
  const char *to_str(std::span<char, TO_STR_BUFFER_SIZE> buffer) const {
    snprintf(buffer.data(), buffer.size(),
             "Timer(id=%s, name=%s, total_seconds=%" PRIu32 ", seconds_left=%" PRIu32 ", is_active=%s)",
             this->id.c_str(), this->name.c_str(), this->total_seconds, this->seconds_left, YESNO(this->is_active));
    return buffer.data();
  }
};
 
struct WakeWord {
  std::string id;
  std::string wake_word;
  std::vector<std::string> trained_languages;
};
 
struct Configuration {
  std::vector<WakeWord> available_wake_words;
  std::vector<std::string> active_wake_words;
  uint32_t max_active_wake_words;
};
 
#ifdef USE_MEDIA_PLAYER
enum class MediaPlayerResponseState : uint8_t {
  IDLE,
  URL_SENT,
  PLAYING,
  FINISHED,
};
#endif
 
class VoiceAssistant final : public Component {
 public:
  VoiceAssistant();
 
  void loop() override;
  void setup() override;
  float get_setup_priority() const override;
  void start_streaming();
  void start_streaming(struct sockaddr_storage *addr, uint16_t port);
  void failed_to_start();
 
  void set_microphone_source(microphone::MicrophoneSource *mic_source) { this->mic_source_ = mic_source; }
  void set_microphone_source2(microphone::MicrophoneSource *mic_source2) { this->mic_source2_ = mic_source2; }
#ifdef USE_MICRO_WAKE_WORD
  void set_micro_wake_word(micro_wake_word::MicroWakeWord *mww) { this->micro_wake_word_ = mww; }
#endif
#ifdef USE_SPEAKER
  void set_speaker(speaker::Speaker *speaker) {
    this->speaker_ = speaker;
    this->local_output_ = true;
  }
#endif
#ifdef USE_MEDIA_PLAYER
  void set_media_player(media_player::MediaPlayer *media_player) {
    this->media_player_ = media_player;
    this->local_output_ = true;
  }
#endif
 
  uint32_t get_legacy_version() const {
#ifdef USE_SPEAKER
    if (this->speaker_ != nullptr) {
      return LEGACY_SPEAKER_SUPPORT;
    }
#endif
    return LEGACY_INITIAL_VERSION;
  }
 
  uint32_t get_feature_flags() const {
    uint32_t flags = 0;
    flags |= VoiceAssistantFeature::FEATURE_VOICE_ASSISTANT;
    flags |= VoiceAssistantFeature::FEATURE_API_AUDIO;
    if (this->mic_source2_ != nullptr) {
      flags |= VoiceAssistantFeature::FEATURE_MULTI_CHANNEL_AUDIO;
    }
#ifdef USE_SPEAKER
    if (this->speaker_ != nullptr) {
      flags |= VoiceAssistantFeature::FEATURE_SPEAKER;
    }
#endif
 
    if (this->has_timers_) {
      flags |= VoiceAssistantFeature::FEATURE_TIMERS;
    }
 
#ifdef USE_MEDIA_PLAYER
    if (this->media_player_ != nullptr) {
      flags |= VoiceAssistantFeature::FEATURE_ANNOUNCE;
      flags |= VoiceAssistantFeature::FEATURE_START_CONVERSATION;
    }
#endif
 
    return flags;
  }
 
  void request_start(bool continuous, bool silence_detection);
  void request_stop();
 
  void on_event(const api::VoiceAssistantEventResponse &msg);
  void on_audio(const api::VoiceAssistantAudio &msg);
  void on_timer_event(const api::VoiceAssistantTimerEventResponse &msg);
  void on_announce(const api::VoiceAssistantAnnounceRequest &msg);
  void on_set_configuration(const std::vector<std::string> &active_wake_words);
  const Configuration &get_configuration();
 
  bool is_running() const { return this->state_ != State::IDLE; }
  void set_continuous(bool continuous) { this->continuous_ = continuous; }
  bool is_continuous() const { return this->continuous_; }
  bool is_running_or_continuous() const { return this->is_running() || this->is_continuous(); }
 
  void set_use_wake_word(bool use_wake_word) { this->use_wake_word_ = use_wake_word; }
 
  void set_noise_suppression_level(uint8_t noise_suppression_level) {
    this->noise_suppression_level_ = noise_suppression_level;
  }
  void set_auto_gain(uint8_t auto_gain) { this->auto_gain_ = auto_gain; }
  void set_volume_multiplier(float volume_multiplier) { this->volume_multiplier_ = volume_multiplier; }
  void set_conversation_timeout(uint32_t conversation_timeout) { this->conversation_timeout_ = conversation_timeout; }
  void set_tts_playback_start_timeout(uint32_t timeout) { this->tts_playback_start_timeout_ = timeout; }
  void reset_conversation_id();
 
  Trigger<> *get_intent_end_trigger() { return &this->intent_end_trigger_; }
  Trigger<> *get_intent_start_trigger() { return &this->intent_start_trigger_; }
  Trigger<std::string> *get_intent_progress_trigger() { return &this->intent_progress_trigger_; }
  Trigger<> *get_listening_trigger() { return &this->listening_trigger_; }
  Trigger<> *get_end_trigger() { return &this->end_trigger_; }
  Trigger<> *get_start_trigger() { return &this->start_trigger_; }
  Trigger<> *get_stt_vad_end_trigger() { return &this->stt_vad_end_trigger_; }
  Trigger<> *get_stt_vad_start_trigger() { return &this->stt_vad_start_trigger_; }
#ifdef USE_SPEAKER
  Trigger<> *get_tts_stream_start_trigger() { return &this->tts_stream_start_trigger_; }
  Trigger<> *get_tts_stream_end_trigger() { return &this->tts_stream_end_trigger_; }
#endif
  Trigger<> *get_wake_word_detected_trigger() { return &this->wake_word_detected_trigger_; }
  Trigger<std::string> *get_stt_end_trigger() { return &this->stt_end_trigger_; }
  Trigger<std::string> *get_tts_end_trigger() { return &this->tts_end_trigger_; }
  Trigger<std::string> *get_tts_start_trigger() { return &this->tts_start_trigger_; }
  Trigger<std::string, std::string> *get_error_trigger() { return &this->error_trigger_; }
  Trigger<> *get_idle_trigger() { return &this->idle_trigger_; }
 
  Trigger<> *get_client_connected_trigger() { return &this->client_connected_trigger_; }
  Trigger<> *get_client_disconnected_trigger() { return &this->client_disconnected_trigger_; }
 
  void client_subscription(api::APIConnection *client, bool subscribe);
  api::APIConnection *get_api_connection() const { return this->api_client_; }
 
  void set_wake_word(const std::string &wake_word) { this->wake_word_ = wake_word; }
 
  Trigger<Timer> *get_timer_started_trigger() { return &this->timer_started_trigger_; }
  Trigger<Timer> *get_timer_updated_trigger() { return &this->timer_updated_trigger_; }
  Trigger<Timer> *get_timer_cancelled_trigger() { return &this->timer_cancelled_trigger_; }
  Trigger<Timer> *get_timer_finished_trigger() { return &this->timer_finished_trigger_; }
  Trigger<const std::vector<Timer> &> *get_timer_tick_trigger() { return &this->timer_tick_trigger_; }
  void set_has_timers(bool has_timers) { this->has_timers_ = has_timers; }
  const std::vector<Timer> &get_timers() const { return this->timers_; }
 
 protected:
  bool allocate_buffers_();
  void clear_buffers_();
  void deallocate_buffers_();
 
  void set_state_(State state);
  void set_state_(State state, State desired_state);
  void set_state_after_mic_stop_(State state);
  void cancel_response_();
  bool start_udp_socket_();
  void signal_stop_();
  void start_playback_timeout_();
 
  // Drains the exposed microphone audio and sends it to Home Assistant over the API in one loop() pass.
  void stream_api_audio_();
  // Handles a pass where at least one configured channel has no audio exposed, timing out a channel that
  // stalls. See audio_channel_stall_start_.
  void handle_channel_stall_(size_t available, size_t available2);
 
  std::unique_ptr<socket::Socket> socket_ = nullptr;
  struct sockaddr_storage dest_addr_;
 
  Trigger<> intent_end_trigger_;
  Trigger<> intent_start_trigger_;
  Trigger<> listening_trigger_;
  Trigger<> end_trigger_;
  Trigger<> start_trigger_;
  Trigger<> stt_vad_start_trigger_;
  Trigger<> stt_vad_end_trigger_;
#ifdef USE_SPEAKER
  Trigger<> tts_stream_start_trigger_;
  Trigger<> tts_stream_end_trigger_;
#endif
  Trigger<std::string> intent_progress_trigger_;
  Trigger<> wake_word_detected_trigger_;
  Trigger<std::string> stt_end_trigger_;
  Trigger<std::string> tts_end_trigger_;
  Trigger<std::string> tts_start_trigger_;
  Trigger<std::string, std::string> error_trigger_;
  Trigger<> idle_trigger_;
 
  Trigger<> client_connected_trigger_;
  Trigger<> client_disconnected_trigger_;
 
  api::APIConnection *api_client_{nullptr};
 
  std::vector<Timer> timers_;
  void timer_tick_();
  Trigger<Timer> timer_started_trigger_;
  Trigger<Timer> timer_finished_trigger_;
  Trigger<Timer> timer_updated_trigger_;
  Trigger<Timer> timer_cancelled_trigger_;
  Trigger<const std::vector<Timer> &> timer_tick_trigger_;
 
  microphone::MicrophoneSource *mic_source_{nullptr};
  microphone::MicrophoneSource *mic_source2_{nullptr};
#ifdef USE_SPEAKER
  void write_speaker_();
  speaker::Speaker *speaker_{nullptr};
  uint8_t *speaker_buffer_{nullptr};
  size_t speaker_buffer_index_{0};
  size_t speaker_buffer_size_{0};
  size_t speaker_bytes_received_{0};
#endif
#ifdef USE_MEDIA_PLAYER
  media_player::MediaPlayer *media_player_{nullptr};
  std::string tts_response_url_;
#endif
 
  std::string conversation_id_;
 
  std::string wake_word_;
 
  // Zero-copy sources that read directly from each microphone channel's ring buffer internal storage.
  // Each source owns its ring buffer; the matching ``ring_buffer_``/``ring_buffer2_`` weak_ptr is used by
  // the microphone callback (a different thread) to write into it.
  std::unique_ptr<audio::RingBufferAudioSource> audio_source_;
  std::unique_ptr<audio::RingBufferAudioSource> audio_source2_;
  std::weak_ptr<ring_buffer::RingBuffer> ring_buffer_;
  std::weak_ptr<ring_buffer::RingBuffer> ring_buffer2_;
 
  // When streaming multiple channels, the send loop holds an exposed chunk on one channel until the other
  // channel also has audio so the channels are always sent together (an empty payload looks like
  // end-of-stream to Home Assistant). Home Assistant has no stream timeout, so a channel that stops
  // producing entirely would hang streaming forever. This records when such an imbalance began so a
  // prolonged one can be detected and stopped; 0 means no imbalance is currently being timed.
  uint32_t audio_channel_stall_start_{0};
 
  float volume_multiplier_;
  uint32_t conversation_timeout_;
  // Keep in sync with TTS_PLAYBACK_START_TIMEOUT_DEFAULT in __init__.py.
  uint32_t tts_playback_start_timeout_{2000};
 
  Configuration config_{};
 
#ifdef USE_MICRO_WAKE_WORD
  micro_wake_word::MicroWakeWord *micro_wake_word_{nullptr};
#endif
 
  // 1 byte members grouped at the end so they share padding instead of each taking a word
  State state_{State::IDLE};
  State desired_state_{State::IDLE};
  AudioMode audio_mode_{AUDIO_MODE_UDP};
  bool has_timers_{false};
  bool timer_tick_running_{false};
#ifdef USE_SPEAKER
  bool wait_for_stream_end_{false};
  bool stream_ended_{false};
#endif
#ifdef USE_MEDIA_PLAYER
  MediaPlayerResponseState media_player_response_state_{MediaPlayerResponseState::IDLE};
  bool started_streaming_tts_{false};
#endif
  bool local_output_{false};
  bool use_wake_word_;
  uint8_t noise_suppression_level_;
  uint8_t auto_gain_;
  bool continuous_{false};
  bool silence_detection_;
  bool continue_conversation_{false};
  bool udp_socket_running_{false};
};
 
extern VoiceAssistant *global_voice_assistant;  // NOLINT(cppcoreguidelines-avoid-non-const-global-variables)
 
}  // namespace esphome::voice_assistant
 
#endif  // USE_VOICE_ASSISTANT