feat(esp-box): Improve audio configuration and behavior (#551)

* feat(esp-box): Improve audio configuration

* always write the full buffer

Author: finger563

components/esp-box/include/esp-box.hpp

@@ -521,11 +521,10 @@ class EspBox : public BaseComponent {
   std::unique_ptr<espp::Task> audio_task_{nullptr};
   // i2s / low-level audio
   i2s_chan_handle_t audio_tx_handle{nullptr};
+  i2s_chan_handle_t audio_rx_handle{nullptr};
   std::vector<uint8_t> audio_tx_buffer;
   StreamBufferHandle_t audio_tx_stream;
   i2s_std_config_t audio_std_cfg;
-  i2s_event_callbacks_t audio_tx_callbacks_;
-  std::atomic<bool> has_sound{false};
 
   // IMU
   std::shared_ptr<Imu> imu_;

components/esp-box/src/audio.cpp

@@ -6,15 +6,6 @@ using namespace espp;
 // Audio Functions   //
 ////////////////////////
 
-static TaskHandle_t play_audio_task_handle_ = NULL;
-
-static bool IRAM_ATTR audio_tx_sent_callback(i2s_chan_handle_t handle, i2s_event_data_t *event,
-                                             void *user_ctx) {
-  // notify the main task that we're done
-  vTaskNotifyGiveFromISR(play_audio_task_handle_, NULL);
-  return true;
-}
-
 bool EspBox::initialize_codec() {
   logger_.info("initializing codec");
 
@@ -27,15 +18,15 @@ bool EspBox::initialize_codec() {
   audio_hal_codec_config_t cfg;
   memset(&cfg, 0, sizeof(cfg));
   cfg.codec_mode = AUDIO_HAL_CODEC_MODE_DECODE;
-  cfg.dac_output = AUDIO_HAL_DAC_OUTPUT_LINE1;
+  cfg.dac_output = AUDIO_HAL_DAC_OUTPUT_ALL; // Enable both L and R outputs
   cfg.i2s_iface.bits = AUDIO_HAL_BIT_LENGTH_16BITS;
   cfg.i2s_iface.fmt = AUDIO_HAL_I2S_NORMAL;
   cfg.i2s_iface.mode = AUDIO_HAL_MODE_SLAVE;
-  cfg.i2s_iface.samples = AUDIO_HAL_16K_SAMPLES;
+  cfg.i2s_iface.samples = AUDIO_HAL_48K_SAMPLES;
 
   ret_val |= es8311_codec_init(&cfg);
-  ret_val |= es8311_set_bits_per_sample(cfg.i2s_iface.bits);
-  ret_val |= es8311_config_fmt((es_i2s_fmt_t)cfg.i2s_iface.fmt);
+  ret_val |= es8311_config_fmt(ES_I2S_NORMAL);
+  ret_val |= es8311_set_bits_per_sample(AUDIO_HAL_BIT_LENGTH_16BITS);
   ret_val |= es8311_codec_set_voice_volume(volume_);
   ret_val |= es8311_codec_ctrl_state(cfg.codec_mode, AUDIO_HAL_CTRL_START);
 
@@ -51,45 +42,25 @@ bool EspBox::initialize_codec() {
 bool EspBox::initialize_i2s(uint32_t default_audio_rate) {
   logger_.info("initializing i2s driver");
   logger_.debug("Using newer I2S standard");
-  i2s_chan_config_t chan_cfg = {
-      .id = i2s_port,
-      .role = I2S_ROLE_MASTER,
-      .dma_desc_num = 16,  // TODO: calculate form audio rate
-      .dma_frame_num = 48, // TODO: calculate from audio rate
-      .auto_clear = true,
-      .auto_clear_before_cb = false,
-      .allow_pd = false,
-      .intr_priority = 0,
-  };
 
-  ESP_ERROR_CHECK(i2s_new_channel(&chan_cfg, &audio_tx_handle, nullptr));
+  i2s_chan_config_t chan_cfg = I2S_CHANNEL_DEFAULT_CONFIG(i2s_port, I2S_ROLE_MASTER);
+  chan_cfg.auto_clear = true; // Auto clear the legacy data in the DMA buffer
+  // chan_cfg.dma_desc_num = 16;
+  // chan_cfg.dma_frame_num = 48;
+  ESP_ERROR_CHECK(i2s_new_channel(&chan_cfg, &audio_tx_handle, &audio_rx_handle));
 
   audio_std_cfg = {
-      .clk_cfg =
-          {
-              .sample_rate_hz = default_audio_rate,
-              .clk_src = I2S_CLK_SRC_DEFAULT,
-              .ext_clk_freq_hz = 0,
-              .mclk_multiple = I2S_MCLK_MULTIPLE_256,
-          },
+      .clk_cfg = I2S_STD_CLK_DEFAULT_CONFIG(default_audio_rate),
       .slot_cfg =
           I2S_STD_PHILIPS_SLOT_DEFAULT_CONFIG(I2S_DATA_BIT_WIDTH_16BIT, I2S_SLOT_MODE_STEREO),
-      .gpio_cfg =
-          {
-              .mclk = i2s_mck_io,
-              .bclk = i2s_bck_io,
-              .ws = i2s_ws_io,
-              .dout = i2s_do_io,
-              .din = i2s_di_io,
-              .invert_flags =
-                  {
-                      .mclk_inv = false,
-                      .bclk_inv = false,
-                      .ws_inv = false,
-                  },
-          },
+      .gpio_cfg = {.mclk = i2s_mck_io,
+                   .bclk = i2s_bck_io,
+                   .ws = i2s_ws_io,
+                   .dout = i2s_do_io,
+                   .din = i2s_di_io,
+                   .invert_flags = {.mclk_inv = false, .bclk_inv = false, .ws_inv = false}},
   };
-  audio_std_cfg.clk_cfg.mclk_multiple = I2S_MCLK_MULTIPLE_256;
+  // audio_std_cfg.clk_cfg.mclk_multiple = I2S_MCLK_MULTIPLE_256;
 
   ESP_ERROR_CHECK(i2s_channel_init_std_mode(audio_tx_handle, &audio_std_cfg));
 
@@ -98,12 +69,6 @@ bool EspBox::initialize_i2s(uint32_t default_audio_rate) {
 
   audio_tx_stream = xStreamBufferCreate(buffer_size * 4, 0);
 
-  play_audio_task_handle_ = xTaskGetCurrentTaskHandle();
-
-  memset(&audio_tx_callbacks_, 0, sizeof(audio_tx_callbacks_));
-  audio_tx_callbacks_.on_sent = audio_tx_sent_callback;
-  i2s_channel_register_event_callback(audio_tx_handle, &audio_tx_callbacks_, NULL);
-
   xStreamBufferReset(audio_tx_stream);
 
   ESP_ERROR_CHECK(i2s_channel_enable(audio_tx_handle));
@@ -155,7 +120,7 @@ bool EspBox::audio_task_callback(std::mutex &m, std::condition_variable &cv, boo
     i2s_channel_write(audio_tx_handle, buffer, buffer_size, NULL, portMAX_DELAY);
   } else {
     xStreamBufferReceive(audio_tx_stream, buffer, available, 0);
-    i2s_channel_write(audio_tx_handle, buffer, available, NULL, portMAX_DELAY);
+    i2s_channel_write(audio_tx_handle, buffer, buffer_size, NULL, portMAX_DELAY);
   }
   return false; // don't stop the task
 }
@@ -194,6 +159,10 @@ void EspBox::audio_sample_rate(uint32_t sample_rate) {
   // stop the channel
   i2s_channel_disable(audio_tx_handle);
   // update the sample rate
+  auto err = es8311_codec_set_sample_rate(sample_rate);
+  if (err != ESP_OK) {
+    logger_.error("Could not set codec sample rate: {}", err);
+  }
   audio_std_cfg.clk_cfg.sample_rate_hz = sample_rate;
   i2s_channel_reconfig_std_clock(audio_tx_handle, &audio_std_cfg.clk_cfg);
   // clear the buffer
@@ -205,11 +174,6 @@ void EspBox::audio_sample_rate(uint32_t sample_rate) {
 void EspBox::play_audio(const std::vector<uint8_t> &data) { play_audio(data.data(), data.size()); }
 
 void EspBox::play_audio(const uint8_t *data, uint32_t num_bytes) {
-  play_audio_task_handle_ = xTaskGetCurrentTaskHandle();
-  if (has_sound) {
-    ulTaskNotifyTake(pdTRUE, portMAX_DELAY);
-  }
   // don't block here
   xStreamBufferSendFromISR(audio_tx_stream, data, num_bytes, NULL);
-  has_sound = true;
 }