diff --git a/esphome/components/i2s_audio/__init__.py b/esphome/components/i2s_audio/__init__.py index 9a2aa0362f..575b914605 100644 --- a/esphome/components/i2s_audio/__init__.py +++ b/esphome/components/i2s_audio/__init__.py @@ -1,6 +1,6 @@ from esphome import pins import esphome.codegen as cg -from esphome.components.esp32 import get_esp32_variant +from esphome.components.esp32 import add_idf_sdkconfig_option, get_esp32_variant from esphome.components.esp32.const import ( VARIANT_ESP32, VARIANT_ESP32C3, @@ -258,6 +258,10 @@ async def to_code(config): if use_legacy(): cg.add_define("USE_I2S_LEGACY") + # Helps avoid callbacks being skipped due to processor load + if CORE.using_esp_idf: + add_idf_sdkconfig_option("CONFIG_I2S_ISR_IRAM_SAFE", True) + cg.add(var.set_lrclk_pin(config[CONF_I2S_LRCLK_PIN])) if CONF_I2S_BCLK_PIN in config: cg.add(var.set_bclk_pin(config[CONF_I2S_BCLK_PIN])) diff --git a/esphome/components/i2s_audio/speaker/i2s_audio_speaker.cpp b/esphome/components/i2s_audio/speaker/i2s_audio_speaker.cpp index 1042a7ebee..6f8c13fe74 100644 --- a/esphome/components/i2s_audio/speaker/i2s_audio_speaker.cpp +++ b/esphome/components/i2s_audio/speaker/i2s_audio_speaker.cpp @@ -9,6 +9,7 @@ #endif #include "esphome/components/audio/audio.h" +#include "esphome/components/audio/audio_transfer_buffer.h" #include "esphome/core/application.h" #include "esphome/core/hal.h" @@ -19,72 +20,33 @@ namespace esphome { namespace i2s_audio { -static const uint8_t DMA_BUFFER_DURATION_MS = 15; +static const uint32_t DMA_BUFFER_DURATION_MS = 15; static const size_t DMA_BUFFERS_COUNT = 4; -static const size_t TASK_DELAY_MS = DMA_BUFFER_DURATION_MS * DMA_BUFFERS_COUNT / 2; - static const size_t TASK_STACK_SIZE = 4096; -static const ssize_t TASK_PRIORITY = 23; +static const ssize_t TASK_PRIORITY = 19; static const size_t I2S_EVENT_QUEUE_COUNT = DMA_BUFFERS_COUNT + 1; static const char *const TAG = "i2s_audio.speaker"; enum SpeakerEventGroupBits : uint32_t { - COMMAND_START = (1 << 0), // starts the speaker task + COMMAND_START = (1 << 0), // indicates loop should start speaker task COMMAND_STOP = (1 << 1), // stops the speaker task COMMAND_STOP_GRACEFULLY = (1 << 2), // Stops the speaker task once all data has been written - STATE_STARTING = (1 << 10), - STATE_RUNNING = (1 << 11), - STATE_STOPPING = (1 << 12), - STATE_STOPPED = (1 << 13), - ERR_TASK_FAILED_TO_START = (1 << 14), - ERR_ESP_INVALID_STATE = (1 << 15), - ERR_ESP_NOT_SUPPORTED = (1 << 16), - ERR_ESP_INVALID_ARG = (1 << 17), - ERR_ESP_INVALID_SIZE = (1 << 18), + + TASK_STARTING = (1 << 10), + TASK_RUNNING = (1 << 11), + TASK_STOPPING = (1 << 12), + TASK_STOPPED = (1 << 13), + ERR_ESP_NO_MEM = (1 << 19), - ERR_ESP_FAIL = (1 << 20), - ALL_ERR_ESP_BITS = ERR_ESP_INVALID_STATE | ERR_ESP_NOT_SUPPORTED | ERR_ESP_INVALID_ARG | ERR_ESP_INVALID_SIZE | - ERR_ESP_NO_MEM | ERR_ESP_FAIL, + + WARN_DROPPED_EVENT = (1 << 20), + ALL_BITS = 0x00FFFFFF, // All valid FreeRTOS event group bits }; -// Translates a SpeakerEventGroupBits ERR_ESP bit to the coressponding esp_err_t -static esp_err_t err_bit_to_esp_err(uint32_t bit) { - switch (bit) { - case SpeakerEventGroupBits::ERR_ESP_INVALID_STATE: - return ESP_ERR_INVALID_STATE; - case SpeakerEventGroupBits::ERR_ESP_INVALID_ARG: - return ESP_ERR_INVALID_ARG; - case SpeakerEventGroupBits::ERR_ESP_INVALID_SIZE: - return ESP_ERR_INVALID_SIZE; - case SpeakerEventGroupBits::ERR_ESP_NO_MEM: - return ESP_ERR_NO_MEM; - case SpeakerEventGroupBits::ERR_ESP_NOT_SUPPORTED: - return ESP_ERR_NOT_SUPPORTED; - default: - return ESP_FAIL; - } -} - -/// @brief Multiplies the input array of Q15 numbers by a Q15 constant factor -/// -/// Based on `dsps_mulc_s16_ansi` from the esp-dsp library: -/// https://github.com/espressif/esp-dsp/blob/master/modules/math/mulc/fixed/dsps_mulc_s16_ansi.c -/// (accessed on 2024-09-30). -/// @param input Array of Q15 numbers -/// @param output Array of Q15 numbers -/// @param len Length of array -/// @param c Q15 constant factor -static void q15_multiplication(const int16_t *input, int16_t *output, size_t len, int16_t c) { - for (int i = 0; i < len; i++) { - int32_t acc = (int32_t) input[i] * (int32_t) c; - output[i] = (int16_t) (acc >> 15); - } -} - // Lists the Q15 fixed point scaling factor for volume reduction. // Has 100 values representing silence and a reduction [49, 48.5, ... 0.5, 0] dB. // dB to PCM scaling factor formula: floating_point_scale_factor = 2^(-db/6.014) @@ -132,51 +94,80 @@ void I2SAudioSpeaker::dump_config() { void I2SAudioSpeaker::loop() { uint32_t event_group_bits = xEventGroupGetBits(this->event_group_); - if (event_group_bits & SpeakerEventGroupBits::STATE_STARTING) { - ESP_LOGD(TAG, "Starting"); + if ((event_group_bits & SpeakerEventGroupBits::COMMAND_START) && (this->state_ == speaker::STATE_STOPPED)) { this->state_ = speaker::STATE_STARTING; - xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::STATE_STARTING); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::COMMAND_START); } - if (event_group_bits & SpeakerEventGroupBits::STATE_RUNNING) { + + // Handle the task's state + if (event_group_bits & SpeakerEventGroupBits::TASK_STARTING) { + ESP_LOGD(TAG, "Starting"); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::TASK_STARTING); + } + if (event_group_bits & SpeakerEventGroupBits::TASK_RUNNING) { ESP_LOGD(TAG, "Started"); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::TASK_RUNNING); this->state_ = speaker::STATE_RUNNING; - xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::STATE_RUNNING); - this->status_clear_warning(); - this->status_clear_error(); } - if (event_group_bits & SpeakerEventGroupBits::STATE_STOPPING) { + if (event_group_bits & SpeakerEventGroupBits::TASK_STOPPING) { ESP_LOGD(TAG, "Stopping"); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::TASK_STOPPING); this->state_ = speaker::STATE_STOPPING; - xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::STATE_STOPPING); } - if (event_group_bits & SpeakerEventGroupBits::STATE_STOPPED) { - if (!this->task_created_) { - ESP_LOGD(TAG, "Stopped"); - this->state_ = speaker::STATE_STOPPED; - xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::ALL_BITS); - this->speaker_task_handle_ = nullptr; - } + if (event_group_bits & SpeakerEventGroupBits::TASK_STOPPED) { + ESP_LOGD(TAG, "Stopped"); + + vTaskDelete(this->speaker_task_handle_); + this->speaker_task_handle_ = nullptr; + + this->stop_i2s_driver_(); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::ALL_BITS); + this->status_clear_error(); + + this->state_ = speaker::STATE_STOPPED; } - if (event_group_bits & SpeakerEventGroupBits::ERR_TASK_FAILED_TO_START) { - this->status_set_error("Failed to start task"); - xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::ERR_TASK_FAILED_TO_START); + // Log any errors encounted by the task + if (event_group_bits & SpeakerEventGroupBits::ERR_ESP_NO_MEM) { + ESP_LOGE(TAG, "Not enough memory"); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_NO_MEM); } - if (event_group_bits & SpeakerEventGroupBits::ALL_ERR_ESP_BITS) { - uint32_t error_bits = event_group_bits & SpeakerEventGroupBits::ALL_ERR_ESP_BITS; - ESP_LOGW(TAG, "Writing failed: %s", esp_err_to_name(err_bit_to_esp_err(error_bits))); - this->status_set_warning(); + // Warn if any playback timestamp events are dropped, which drastically reduces synced playback accuracy + if (event_group_bits & SpeakerEventGroupBits::WARN_DROPPED_EVENT) { + ESP_LOGW(TAG, "Event dropped, synchronized playback accuracy is reduced"); + xEventGroupClearBits(this->event_group_, SpeakerEventGroupBits::WARN_DROPPED_EVENT); } - if (event_group_bits & SpeakerEventGroupBits::ERR_ESP_NOT_SUPPORTED) { - this->status_set_error("Failed to adjust bus to match incoming audio"); - ESP_LOGE(TAG, "Incompatible audio format: sample rate = %" PRIu32 ", channels = %u, bits per sample = %u", - this->audio_stream_info_.get_sample_rate(), this->audio_stream_info_.get_channels(), - this->audio_stream_info_.get_bits_per_sample()); - } + // Handle the speaker's state + switch (this->state_) { + case speaker::STATE_STARTING: + if (this->status_has_error()) { + break; + } - xEventGroupClearBits(this->event_group_, ALL_ERR_ESP_BITS); + if (this->start_i2s_driver_(this->audio_stream_info_) != ESP_OK) { + ESP_LOGE(TAG, "Driver failed to start; retrying in 1 second"); + this->status_momentary_error("driver-faiure", 1000); + break; + } + + if (this->speaker_task_handle_ == nullptr) { + xTaskCreate(I2SAudioSpeaker::speaker_task, "speaker_task", TASK_STACK_SIZE, (void *) this, TASK_PRIORITY, + &this->speaker_task_handle_); + + if (this->speaker_task_handle_ == nullptr) { + ESP_LOGE(TAG, "Task failed to start, retrying in 1 second"); + this->status_momentary_error("task-failure", 1000); + this->stop_i2s_driver_(); // Stops the driver to return the lock; will be reloaded in next attempt + } + } + break; + case speaker::STATE_RUNNING: // Intentional fallthrough + case speaker::STATE_STOPPING: // Intentional fallthrough + case speaker::STATE_STOPPED: + break; + } } void I2SAudioSpeaker::set_volume(float volume) { @@ -227,83 +218,76 @@ size_t I2SAudioSpeaker::play(const uint8_t *data, size_t length, TickType_t tick this->start(); } - if ((this->state_ != speaker::STATE_RUNNING) || (this->audio_ring_buffer_.use_count() != 1)) { + if (this->state_ != speaker::STATE_RUNNING) { // Unable to write data to a running speaker, so delay the max amount of time so it can get ready vTaskDelay(ticks_to_wait); ticks_to_wait = 0; } size_t bytes_written = 0; - if ((this->state_ == speaker::STATE_RUNNING) && (this->audio_ring_buffer_.use_count() == 1)) { - // Only one owner of the ring buffer (the speaker task), so the ring buffer is allocated and no other components are - // attempting to write to it. - - // Temporarily share ownership of the ring buffer so it won't be deallocated while writing - std::shared_ptr temp_ring_buffer = this->audio_ring_buffer_; - bytes_written = temp_ring_buffer->write_without_replacement((void *) data, length, ticks_to_wait); + if (this->state_ == speaker::STATE_RUNNING) { + std::shared_ptr temp_ring_buffer = this->audio_ring_buffer_.lock(); + if (temp_ring_buffer.use_count() == 2) { + // Only the speaker task and this temp_ring_buffer own the ring buffer, so its safe to write to + bytes_written = temp_ring_buffer->write_without_replacement((void *) data, length, ticks_to_wait); + } } return bytes_written; } bool I2SAudioSpeaker::has_buffered_data() const { - if (this->audio_ring_buffer_ != nullptr) { - return this->audio_ring_buffer_->available() > 0; + if (this->audio_ring_buffer_.use_count() > 0) { + std::shared_ptr temp_ring_buffer = this->audio_ring_buffer_.lock(); + return temp_ring_buffer->available() > 0; } return false; } void I2SAudioSpeaker::speaker_task(void *params) { I2SAudioSpeaker *this_speaker = (I2SAudioSpeaker *) params; - this_speaker->task_created_ = true; - uint32_t event_group_bits = - xEventGroupWaitBits(this_speaker->event_group_, - SpeakerEventGroupBits::COMMAND_START | SpeakerEventGroupBits::COMMAND_STOP | - SpeakerEventGroupBits::COMMAND_STOP_GRACEFULLY, // Bit message to read - pdTRUE, // Clear the bits on exit - pdFALSE, // Don't wait for all the bits, - portMAX_DELAY); // Block indefinitely until a bit is set - - if (event_group_bits & (SpeakerEventGroupBits::COMMAND_STOP | SpeakerEventGroupBits::COMMAND_STOP_GRACEFULLY)) { - // Received a stop signal before the task was requested to start - this_speaker->delete_task_(0); - } - - xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::STATE_STARTING); - - audio::AudioStreamInfo audio_stream_info = this_speaker->audio_stream_info_; + xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::TASK_STARTING); const uint32_t dma_buffers_duration_ms = DMA_BUFFER_DURATION_MS * DMA_BUFFERS_COUNT; // Ensure ring buffer duration is at least the duration of all DMA buffers const uint32_t ring_buffer_duration = std::max(dma_buffers_duration_ms, this_speaker->buffer_duration_ms_); // The DMA buffers may have more bits per sample, so calculate buffer sizes based in the input audio stream info - const size_t data_buffer_size = audio_stream_info.ms_to_bytes(dma_buffers_duration_ms); - const size_t ring_buffer_size = audio_stream_info.ms_to_bytes(ring_buffer_duration); + const size_t ring_buffer_size = this_speaker->current_stream_info_.ms_to_bytes(ring_buffer_duration); - const size_t single_dma_buffer_input_size = data_buffer_size / DMA_BUFFERS_COUNT; + const uint32_t frames_to_fill_single_dma_buffer = + this_speaker->current_stream_info_.ms_to_frames(DMA_BUFFER_DURATION_MS); + const size_t bytes_to_fill_single_dma_buffer = + this_speaker->current_stream_info_.frames_to_bytes(frames_to_fill_single_dma_buffer); - if (this_speaker->send_esp_err_to_event_group_(this_speaker->allocate_buffers_(data_buffer_size, ring_buffer_size))) { - // Failed to allocate buffers - xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::ERR_ESP_NO_MEM); - this_speaker->delete_task_(data_buffer_size); + bool successful_setup = false; + std::unique_ptr transfer_buffer = + audio::AudioSourceTransferBuffer::create(bytes_to_fill_single_dma_buffer); + + if (transfer_buffer != nullptr) { + std::shared_ptr temp_ring_buffer = RingBuffer::create(ring_buffer_size); + if (temp_ring_buffer.use_count() == 1) { + transfer_buffer->set_source(temp_ring_buffer); + this_speaker->audio_ring_buffer_ = temp_ring_buffer; + successful_setup = true; + } } - if (!this_speaker->send_esp_err_to_event_group_(this_speaker->start_i2s_driver_(audio_stream_info))) { - xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::STATE_RUNNING); - + if (!successful_setup) { + xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::ERR_ESP_NO_MEM); + } else { bool stop_gracefully = false; + bool tx_dma_underflow = true; + + uint32_t frames_written = 0; uint32_t last_data_received_time = millis(); - bool tx_dma_underflow = false; - this_speaker->accumulated_frames_written_ = 0; + xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::TASK_RUNNING); - // Keep looping if paused, there is no timeout configured, or data was received more recently than the configured - // timeout while (this_speaker->pause_state_ || !this_speaker->timeout_.has_value() || (millis() - last_data_received_time) <= this_speaker->timeout_.value()) { - event_group_bits = xEventGroupGetBits(this_speaker->event_group_); + uint32_t event_group_bits = xEventGroupGetBits(this_speaker->event_group_); if (event_group_bits & SpeakerEventGroupBits::COMMAND_STOP) { xEventGroupClearBits(this_speaker->event_group_, SpeakerEventGroupBits::COMMAND_STOP); @@ -314,7 +298,7 @@ void I2SAudioSpeaker::speaker_task(void *params) { stop_gracefully = true; } - if (this_speaker->audio_stream_info_ != audio_stream_info) { + if (this_speaker->audio_stream_info_ != this_speaker->current_stream_info_) { // Audio stream info changed, stop the speaker task so it will restart with the proper settings. break; } @@ -326,36 +310,75 @@ void I2SAudioSpeaker::speaker_task(void *params) { } } #else - bool overflow; - while (xQueueReceive(this_speaker->i2s_event_queue_, &overflow, 0)) { - if (overflow) { + int64_t write_timestamp; + while (xQueueReceive(this_speaker->i2s_event_queue_, &write_timestamp, 0)) { + // Receives timing events from the I2S on_sent callback. If actual audio data was sent in this event, it passes + // on the timing info via the audio_output_callback. + uint32_t frames_sent = frames_to_fill_single_dma_buffer; + if (frames_to_fill_single_dma_buffer > frames_written) { tx_dma_underflow = true; + frames_sent = frames_written; + const uint32_t frames_zeroed = frames_to_fill_single_dma_buffer - frames_written; + write_timestamp -= this_speaker->current_stream_info_.frames_to_microseconds(frames_zeroed); + } else { + tx_dma_underflow = false; + } + frames_written -= frames_sent; + if (frames_sent > 0) { + this_speaker->audio_output_callback_(frames_sent, write_timestamp); } } #endif if (this_speaker->pause_state_) { // Pause state is accessed atomically, so thread safe - // Delay so the task can yields, then skip transferring audio data - delay(TASK_DELAY_MS); + // Delay so the task yields, then skip transferring audio data + vTaskDelay(pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS)); continue; } - size_t bytes_read = this_speaker->audio_ring_buffer_->read((void *) this_speaker->data_buffer_, data_buffer_size, - pdMS_TO_TICKS(TASK_DELAY_MS)); + // Wait half the duration of the data already written to the DMA buffers for new audio data + // The millisecond helper modifies the frames_written variable, so use the microsecond helper and divide by 1000 + const uint32_t read_delay = + (this_speaker->current_stream_info_.frames_to_microseconds(frames_written) / 1000) / 2; + + uint8_t *new_data = transfer_buffer->get_buffer_end(); // track start of any newly copied bytes + size_t bytes_read = transfer_buffer->transfer_data_from_source(pdMS_TO_TICKS(read_delay)); if (bytes_read > 0) { - if ((audio_stream_info.get_bits_per_sample() == 16) && (this_speaker->q15_volume_factor_ < INT16_MAX)) { - // Scale samples by the volume factor in place - q15_multiplication((int16_t *) this_speaker->data_buffer_, (int16_t *) this_speaker->data_buffer_, - bytes_read / sizeof(int16_t), this_speaker->q15_volume_factor_); + if (this_speaker->q15_volume_factor_ < INT16_MAX) { + // Apply the software volume adjustment by unpacking the sample into a Q31 fixed-point number, shifting it, + // multiplying by the volume factor, and packing the sample back into the original bytes per sample. + + const size_t bytes_per_sample = this_speaker->current_stream_info_.samples_to_bytes(1); + const uint32_t len = bytes_read / bytes_per_sample; + + // Use Q16 for samples with 1 or 2 bytes: shifted_sample * gain_factor is Q16 * Q15 -> Q31 + int32_t shift = 15; // Q31 -> Q16 + int32_t gain_factor = this_speaker->q15_volume_factor_; // Q15 + + if (bytes_per_sample >= 3) { + // Use Q23 for samples with 3 or 4 bytes: shifted_sample * gain_factor is Q23 * Q8 -> Q31 + + shift = 8; // Q31 -> Q23 + gain_factor >>= 7; // Q15 -> Q8 + } + + for (uint32_t i = 0; i < len; ++i) { + int32_t sample = + audio::unpack_audio_sample_to_q31(&new_data[i * bytes_per_sample], bytes_per_sample); // Q31 + sample >>= shift; + sample *= gain_factor; // Q31 + audio::pack_q31_as_audio_sample(sample, &new_data[i * bytes_per_sample], bytes_per_sample); + } } #ifdef USE_ESP32_VARIANT_ESP32 // For ESP32 8/16 bit mono mode samples need to be switched. - if (audio_stream_info.get_channels() == 1 && audio_stream_info.get_bits_per_sample() <= 16) { + if (this_speaker->current_stream_info_.get_channels() == 1 && + this_speaker->current_stream_info_.get_bits_per_sample() <= 16) { size_t len = bytes_read / sizeof(int16_t); - int16_t *tmp_buf = (int16_t *) this_speaker->data_buffer_; + int16_t *tmp_buf = (int16_t *) new_data; for (int i = 0; i < len; i += 2) { int16_t tmp = tmp_buf[i]; tmp_buf[i] = tmp_buf[i + 1]; @@ -363,62 +386,87 @@ void I2SAudioSpeaker::speaker_task(void *params) { } } #endif - // Write the audio data to a single DMA buffer at a time to reduce latency for the audio duration played - // callback. - const uint32_t batches = (bytes_read + single_dma_buffer_input_size - 1) / single_dma_buffer_input_size; + } - for (uint32_t i = 0; i < batches; ++i) { - size_t bytes_written = 0; - size_t bytes_to_write = std::min(single_dma_buffer_input_size, bytes_read); - -#ifdef USE_I2S_LEGACY - if (audio_stream_info.get_bits_per_sample() == (uint8_t) this_speaker->bits_per_sample_) { - i2s_write(this_speaker->parent_->get_port(), this_speaker->data_buffer_ + i * single_dma_buffer_input_size, - bytes_to_write, &bytes_written, pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS * 5)); - } else if (audio_stream_info.get_bits_per_sample() < (uint8_t) this_speaker->bits_per_sample_) { - i2s_write_expand(this_speaker->parent_->get_port(), - this_speaker->data_buffer_ + i * single_dma_buffer_input_size, bytes_to_write, - audio_stream_info.get_bits_per_sample(), this_speaker->bits_per_sample_, &bytes_written, - pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS * 5)); - } -#else - i2s_channel_write(this_speaker->tx_handle_, this_speaker->data_buffer_ + i * single_dma_buffer_input_size, - bytes_to_write, &bytes_written, pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS * 5)); -#endif - - int64_t now = esp_timer_get_time(); - - if (bytes_written != bytes_to_write) { - xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::ERR_ESP_INVALID_SIZE); - } - bytes_read -= bytes_written; - - this_speaker->audio_output_callback_(audio_stream_info.bytes_to_frames(bytes_written), - now + dma_buffers_duration_ms * 1000); - - tx_dma_underflow = false; - last_data_received_time = millis(); - } - } else { - // No data received + if (transfer_buffer->available() == 0) { if (stop_gracefully && tx_dma_underflow) { break; } + vTaskDelay(pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS / 2)); + } else { + size_t bytes_written = 0; +#ifdef USE_I2S_LEGACY + if (this_speaker->current_stream_info_.get_bits_per_sample() == (uint8_t) this_speaker->bits_per_sample_) { + i2s_write(this_speaker->parent_->get_port(), transfer_buffer->get_buffer_start(), + transfer_buffer->available(), &bytes_written, pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS)); + } else if (this_speaker->current_stream_info_.get_bits_per_sample() < + (uint8_t) this_speaker->bits_per_sample_) { + i2s_write_expand(this_speaker->parent_->get_port(), transfer_buffer->get_buffer_start(), + transfer_buffer->available(), this_speaker->current_stream_info_.get_bits_per_sample(), + this_speaker->bits_per_sample_, &bytes_written, pdMS_TO_TICKS(DMA_BUFFER_DURATION_MS)); + } +#else + if (tx_dma_underflow) { + // Temporarily disable channel and callback to reset the I2S driver's internal DMA buffer queue so timing + // callbacks are accurate. Preload the data. + i2s_channel_disable(this_speaker->tx_handle_); + const i2s_event_callbacks_t callbacks = { + .on_sent = nullptr, + }; + + i2s_channel_register_event_callback(this_speaker->tx_handle_, &callbacks, this_speaker); + i2s_channel_preload_data(this_speaker->tx_handle_, transfer_buffer->get_buffer_start(), + transfer_buffer->available(), &bytes_written); + } else { + // Audio is already playing, use regular I2S write to add to the DMA buffers + i2s_channel_write(this_speaker->tx_handle_, transfer_buffer->get_buffer_start(), transfer_buffer->available(), + &bytes_written, DMA_BUFFER_DURATION_MS); + } +#endif + if (bytes_written > 0) { + last_data_received_time = millis(); + frames_written += this_speaker->current_stream_info_.bytes_to_frames(bytes_written); + transfer_buffer->decrease_buffer_length(bytes_written); + if (tx_dma_underflow) { + tx_dma_underflow = false; +#ifndef USE_I2S_LEGACY + // Reset the event queue timestamps + // Enable the on_sent callback to accurately track the timestamps of played audio + // Enable the I2S channel to start sending the preloaded audio + + xQueueReset(this_speaker->i2s_event_queue_); + + const i2s_event_callbacks_t callbacks = { + .on_sent = i2s_on_sent_cb, + }; + i2s_channel_register_event_callback(this_speaker->tx_handle_, &callbacks, this_speaker); + + i2s_channel_enable(this_speaker->tx_handle_); +#endif + } +#ifdef USE_I2S_LEGACY + // The legacy driver doesn't easily support the callback approach for timestamps, so fall back to a direct but + // less accurate approach. + this_speaker->audio_output_callback_(this_speaker->current_stream_info_.bytes_to_frames(bytes_written), + esp_timer_get_time() + dma_buffers_duration_ms * 1000); +#endif + } } } - - xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::STATE_STOPPING); -#ifdef USE_I2S_LEGACY - i2s_driver_uninstall(this_speaker->parent_->get_port()); -#else - i2s_channel_disable(this_speaker->tx_handle_); - i2s_del_channel(this_speaker->tx_handle_); -#endif - - this_speaker->parent_->unlock(); } - this_speaker->delete_task_(data_buffer_size); + xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::TASK_STOPPING); + + if (transfer_buffer != nullptr) { + transfer_buffer.reset(); + } + + xEventGroupSetBits(this_speaker->event_group_, SpeakerEventGroupBits::TASK_STOPPED); + + while (true) { + // Continuously delay until the loop method deletes the task + vTaskDelay(pdMS_TO_TICKS(10)); + } } void I2SAudioSpeaker::start() { @@ -427,16 +475,7 @@ void I2SAudioSpeaker::start() { if ((this->state_ == speaker::STATE_STARTING) || (this->state_ == speaker::STATE_RUNNING)) return; - if (!this->task_created_ && (this->speaker_task_handle_ == nullptr)) { - xTaskCreate(I2SAudioSpeaker::speaker_task, "speaker_task", TASK_STACK_SIZE, (void *) this, TASK_PRIORITY, - &this->speaker_task_handle_); - - if (this->speaker_task_handle_ != nullptr) { - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::COMMAND_START); - } else { - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_TASK_FAILED_TO_START); - } - } + xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::COMMAND_START); } void I2SAudioSpeaker::stop() { this->stop_(false); } @@ -456,61 +495,16 @@ void I2SAudioSpeaker::stop_(bool wait_on_empty) { } } -bool I2SAudioSpeaker::send_esp_err_to_event_group_(esp_err_t err) { - switch (err) { - case ESP_OK: - return false; - case ESP_ERR_INVALID_STATE: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_INVALID_STATE); - return true; - case ESP_ERR_INVALID_ARG: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_INVALID_ARG); - return true; - case ESP_ERR_INVALID_SIZE: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_INVALID_SIZE); - return true; - case ESP_ERR_NO_MEM: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_NO_MEM); - return true; - case ESP_ERR_NOT_SUPPORTED: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_NOT_SUPPORTED); - return true; - default: - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::ERR_ESP_FAIL); - return true; - } -} - -esp_err_t I2SAudioSpeaker::allocate_buffers_(size_t data_buffer_size, size_t ring_buffer_size) { - if (this->data_buffer_ == nullptr) { - // Allocate data buffer for temporarily storing audio from the ring buffer before writing to the I2S bus - RAMAllocator allocator; - this->data_buffer_ = allocator.allocate(data_buffer_size); - } - - if (this->data_buffer_ == nullptr) { - return ESP_ERR_NO_MEM; - } - - if (this->audio_ring_buffer_.use_count() == 0) { - // Allocate ring buffer. Uses a shared_ptr to ensure it isn't improperly deallocated. - this->audio_ring_buffer_ = RingBuffer::create(ring_buffer_size); - } - - if (this->audio_ring_buffer_ == nullptr) { - return ESP_ERR_NO_MEM; - } - - return ESP_OK; -} - esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_stream_info) { + this->current_stream_info_ = audio_stream_info; // store the stream info settings the driver will use + #ifdef USE_I2S_LEGACY if ((this->i2s_mode_ & I2S_MODE_SLAVE) && (this->sample_rate_ != audio_stream_info.get_sample_rate())) { // NOLINT #else if ((this->i2s_role_ & I2S_ROLE_SLAVE) && (this->sample_rate_ != audio_stream_info.get_sample_rate())) { // NOLINT #endif // Can't reconfigure I2S bus, so the sample rate must match the configured value + ESP_LOGE(TAG, "Audio stream settings are not compatible with this I2S configuration"); return ESP_ERR_NOT_SUPPORTED; } @@ -521,10 +515,12 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea (i2s_slot_bit_width_t) audio_stream_info.get_bits_per_sample() > this->slot_bit_width_) { #endif // Currently can't handle the case when the incoming audio has more bits per sample than the configured value + ESP_LOGE(TAG, "Audio streams with more bits per sample than the I2S speaker's configuration is not supported"); return ESP_ERR_NOT_SUPPORTED; } if (!this->parent_->try_lock()) { + ESP_LOGE(TAG, "Parent I2S bus not free"); return ESP_ERR_INVALID_STATE; } @@ -575,6 +571,7 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea esp_err_t err = i2s_driver_install(this->parent_->get_port(), &config, I2S_EVENT_QUEUE_COUNT, &this->i2s_event_queue_); if (err != ESP_OK) { + ESP_LOGE(TAG, "Failed to install I2S legacy driver"); // Failed to install the driver, so unlock the I2S port this->parent_->unlock(); return err; @@ -595,6 +592,7 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea if (err != ESP_OK) { // Failed to set the data out pin, so uninstall the driver and unlock the I2S port + ESP_LOGE(TAG, "Failed to set the data out pin"); i2s_driver_uninstall(this->parent_->get_port()); this->parent_->unlock(); } @@ -605,10 +603,12 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea .dma_desc_num = DMA_BUFFERS_COUNT, .dma_frame_num = dma_buffer_length, .auto_clear = true, + .intr_priority = 3, }; /* Allocate a new TX channel and get the handle of this channel */ esp_err_t err = i2s_new_channel(&chan_cfg, &this->tx_handle_, NULL); if (err != ESP_OK) { + ESP_LOGE(TAG, "Failed to allocate new I2S channel"); this->parent_->unlock(); return err; } @@ -652,7 +652,11 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea // per sample causes the audio to play too fast. Setting the ws_width to the configured slot bit width seems to // make it play at the correct speed while sending more bits per slot. if (this->slot_bit_width_ != I2S_SLOT_BIT_WIDTH_AUTO) { - std_slot_cfg.ws_width = static_cast(this->slot_bit_width_); + uint32_t configured_bit_width = static_cast(this->slot_bit_width_); + std_slot_cfg.ws_width = configured_bit_width; + if (configured_bit_width > 16) { + std_slot_cfg.msb_right = false; + } } #else std_slot_cfg.slot_bit_width = this->slot_bit_width_; @@ -670,54 +674,56 @@ esp_err_t I2SAudioSpeaker::start_i2s_driver_(audio::AudioStreamInfo &audio_strea err = i2s_channel_init_std_mode(this->tx_handle_, &std_cfg); if (err != ESP_OK) { + ESP_LOGE(TAG, "Failed to initialize channel"); i2s_del_channel(this->tx_handle_); + this->tx_handle_ = nullptr; this->parent_->unlock(); return err; } if (this->i2s_event_queue_ == nullptr) { - this->i2s_event_queue_ = xQueueCreate(1, sizeof(bool)); + this->i2s_event_queue_ = xQueueCreate(I2S_EVENT_QUEUE_COUNT, sizeof(int64_t)); } - const i2s_event_callbacks_t callbacks = { - .on_send_q_ovf = i2s_overflow_cb, - }; - i2s_channel_register_event_callback(this->tx_handle_, &callbacks, this); - - /* Before reading data, start the TX channel first */ i2s_channel_enable(this->tx_handle_); - if (err != ESP_OK) { - i2s_del_channel(this->tx_handle_); - this->parent_->unlock(); - } #endif return err; } -void I2SAudioSpeaker::delete_task_(size_t buffer_size) { - this->audio_ring_buffer_.reset(); // Releases ownership of the shared_ptr +#ifndef USE_I2S_LEGACY +bool IRAM_ATTR I2SAudioSpeaker::i2s_on_sent_cb(i2s_chan_handle_t handle, i2s_event_data_t *event, void *user_ctx) { + int64_t now = esp_timer_get_time(); - if (this->data_buffer_ != nullptr) { - RAMAllocator allocator; - allocator.deallocate(this->data_buffer_, buffer_size); - this->data_buffer_ = nullptr; + BaseType_t need_yield1 = pdFALSE; + BaseType_t need_yield2 = pdFALSE; + BaseType_t need_yield3 = pdFALSE; + + I2SAudioSpeaker *this_speaker = (I2SAudioSpeaker *) user_ctx; + + if (xQueueIsQueueFullFromISR(this_speaker->i2s_event_queue_)) { + // Queue is full, so discard the oldest event and set the warning flag to inform the user + int64_t dummy; + xQueueReceiveFromISR(this_speaker->i2s_event_queue_, &dummy, &need_yield1); + xEventGroupSetBitsFromISR(this_speaker->event_group_, SpeakerEventGroupBits::WARN_DROPPED_EVENT, &need_yield2); } - xEventGroupSetBits(this->event_group_, SpeakerEventGroupBits::STATE_STOPPED); + xQueueSendToBackFromISR(this_speaker->i2s_event_queue_, &now, &need_yield3); - this->task_created_ = false; - vTaskDelete(nullptr); -} - -#ifndef USE_I2S_LEGACY -bool IRAM_ATTR I2SAudioSpeaker::i2s_overflow_cb(i2s_chan_handle_t handle, i2s_event_data_t *event, void *user_ctx) { - I2SAudioSpeaker *this_speaker = (I2SAudioSpeaker *) user_ctx; - bool overflow = true; - xQueueOverwrite(this_speaker->i2s_event_queue_, &overflow); - return false; + return need_yield1 | need_yield2 | need_yield3; } #endif +void I2SAudioSpeaker::stop_i2s_driver_() { +#ifdef USE_I2S_LEGACY + i2s_driver_uninstall(this->parent_->get_port()); +#else + i2s_channel_disable(this->tx_handle_); + i2s_del_channel(this->tx_handle_); + this->tx_handle_ = nullptr; +#endif + this->parent_->unlock(); +} + } // namespace i2s_audio } // namespace esphome diff --git a/esphome/components/i2s_audio/speaker/i2s_audio_speaker.h b/esphome/components/i2s_audio/speaker/i2s_audio_speaker.h index eb2a0ae756..1d03a4c495 100644 --- a/esphome/components/i2s_audio/speaker/i2s_audio_speaker.h +++ b/esphome/components/i2s_audio/speaker/i2s_audio_speaker.h @@ -72,70 +72,57 @@ class I2SAudioSpeaker : public I2SAudioOut, public speaker::Speaker, public Comp protected: /// @brief Function for the FreeRTOS task handling audio output. - /// After receiving the COMMAND_START signal, allocates space for the buffers, starts the I2S driver, and reads - /// audio from the ring buffer and writes audio to the I2S port. Stops immmiately after receiving the COMMAND_STOP - /// signal and stops only after the ring buffer is empty after receiving the COMMAND_STOP_GRACEFULLY signal. Stops if - /// the ring buffer hasn't read data for more than timeout_ milliseconds. When stopping, it deallocates the buffers, - /// stops the I2S driver, unlocks the I2S port, and deletes the task. It communicates the state and any errors via - /// event_group_. + /// Allocates space for the buffers, reads audio from the ring buffer and writes audio to the I2S port. Stops + /// immmiately after receiving the COMMAND_STOP signal and stops only after the ring buffer is empty after receiving + /// the COMMAND_STOP_GRACEFULLY signal. Stops if the ring buffer hasn't read data for more than timeout_ milliseconds. + /// When stopping, it deallocates the buffers. It communicates its state and any errors via ``event_group_``. /// @param params I2SAudioSpeaker component static void speaker_task(void *params); - /// @brief Sends a stop command to the speaker task via event_group_. + /// @brief Sends a stop command to the speaker task via ``event_group_``. /// @param wait_on_empty If false, sends the COMMAND_STOP signal. If true, sends the COMMAND_STOP_GRACEFULLY signal. void stop_(bool wait_on_empty); - /// @brief Sets the corresponding ERR_ESP event group bits. - /// @param err esp_err_t error code. - /// @return True if an ERR_ESP bit is set and false if err == ESP_OK - bool send_esp_err_to_event_group_(esp_err_t err); - #ifndef USE_I2S_LEGACY - static bool i2s_overflow_cb(i2s_chan_handle_t handle, i2s_event_data_t *event, void *user_ctx); + /// @brief Callback function used to send playback timestamps the to the speaker task. + /// @param handle (i2s_chan_handle_t) + /// @param event (i2s_event_data_t) + /// @param user_ctx (void*) User context pointer that the callback accesses + /// @return True if a higher priority task was interrupted + static bool i2s_on_sent_cb(i2s_chan_handle_t handle, i2s_event_data_t *event, void *user_ctx); #endif - /// @brief Allocates the data buffer and ring buffer - /// @param data_buffer_size Number of bytes to allocate for the data buffer. - /// @param ring_buffer_size Number of bytes to allocate for the ring buffer. - /// @return ESP_ERR_NO_MEM if either buffer fails to allocate - /// ESP_OK if successful - esp_err_t allocate_buffers_(size_t data_buffer_size, size_t ring_buffer_size); - /// @brief Starts the ESP32 I2S driver. /// Attempts to lock the I2S port, starts the I2S driver using the passed in stream information, and sets the data out - /// pin. If it fails, it will unlock the I2S port and uninstall the driver, if necessary. + /// pin. If it fails, it will unlock the I2S port and uninstalls the driver, if necessary. /// @param audio_stream_info Stream information for the I2S driver. /// @return ESP_ERR_NOT_ALLOWED if the I2S port can't play the incoming audio stream. /// ESP_ERR_INVALID_STATE if the I2S port is already locked. - /// ESP_ERR_INVALID_ARG if nstalling the driver or setting the data outpin fails due to a parameter error. + /// ESP_ERR_INVALID_ARG if installing the driver or setting the data outpin fails due to a parameter error. /// ESP_ERR_NO_MEM if the driver fails to install due to a memory allocation error. - /// ESP_FAIL if setting the data out pin fails due to an IO error ESP_OK if successful + /// ESP_FAIL if setting the data out pin fails due to an IO error + /// ESP_OK if successful esp_err_t start_i2s_driver_(audio::AudioStreamInfo &audio_stream_info); - /// @brief Deletes the speaker's task. - /// Deallocates the data_buffer_ and audio_ring_buffer_, if necessary, and deletes the task. Should only be called by - /// the speaker_task itself. - /// @param buffer_size The allocated size of the data_buffer_. - void delete_task_(size_t buffer_size); + /// @brief Stops the I2S driver and unlocks the I2S port + void stop_i2s_driver_(); TaskHandle_t speaker_task_handle_{nullptr}; EventGroupHandle_t event_group_{nullptr}; QueueHandle_t i2s_event_queue_; - uint8_t *data_buffer_; - std::shared_ptr audio_ring_buffer_; + std::weak_ptr audio_ring_buffer_; uint32_t buffer_duration_ms_; optional timeout_; - bool task_created_{false}; bool pause_state_{false}; int16_t q15_volume_factor_{INT16_MAX}; - size_t bytes_written_{0}; + audio::AudioStreamInfo current_stream_info_; // The currently loaded driver's stream info #ifdef USE_I2S_LEGACY #if SOC_I2S_SUPPORTS_DAC @@ -148,8 +135,6 @@ class I2SAudioSpeaker : public I2SAudioOut, public speaker::Speaker, public Comp std::string i2s_comm_fmt_; i2s_chan_handle_t tx_handle_; #endif - - uint32_t accumulated_frames_written_{0}; }; } // namespace i2s_audio