sascha/esphome
1
0
Fork 0
mirror of https://github.com/esphome/esphome.git synced 2025-10-30 22:53:59 +00:00
Code Issues Releases Wiki Activity

[i2s_audio] Use high-pass filter for dc offset correction (#10005)

Browse Source
This commit is contained in:
Chris Beswick
2025-08-04 15:43:44 +01:00
committed by GitHub
parent fbbb791b0d
commit d59476d0e1
2 changed files with 48 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

64
esphome/components/i2s_audio/microphone/i2s_audio_microphone.cpp
View File

@@ -24,9 +24,6 @@ static const uint32_t READ_DURATION_MS = 16;
static const size_t TASK_STACK_SIZE = 4096;
static const ssize_t TASK_PRIORITY = 23;
// Use an exponential moving average to correct a DC offset with weight factor 1/1000
static const int32_t DC_OFFSET_MOVING_AVERAGE_COEFFICIENT_DENOMINATOR = 1000;
static const char *const TAG = "i2s_audio.microphone";
enum MicrophoneEventGroupBits : uint32_t {
@@ -381,26 +378,57 @@ void I2SAudioMicrophone::mic_task(void *params) {
}
void I2SAudioMicrophone::fix_dc_offset_(std::vector<uint8_t> &data) {
/**
* From https://www.musicdsp.org/en/latest/Filters/135-dc-filter.html:
*
* y(n) = x(n) - x(n-1) + R * y(n-1)
* R = 1 - (pi * 2 * frequency / samplerate)
*
* From https://en.wikipedia.org/wiki/Hearing_range:
* The human range is commonly given as 20Hz up.
*
* From https://en.wikipedia.org/wiki/High-resolution_audio:
* A reasonable upper bound for sample rate seems to be 96kHz.
*
* Calculate R value for 20Hz on a 96kHz sample rate:
* R = 1 - (pi * 2 * 20 / 96000)
* R = 0.9986910031
*
* Transform floating point to bit-shifting approximation:
* output = input - prev_input + R * prev_output
* output = input - prev_input + (prev_output - (prev_output >> S))
*
* Approximate bit-shift value S from R:
* R = 1 - (1 >> S)
* R = 1 - (1 / 2^S)
* R = 1 - 2^-S
* 0.9986910031 = 1 - 2^-S
* S = 9.57732 ~= 10
*
* Actual R from S:
* R = 1 - 2^-10 = 0.9990234375
*
* Confirm this has effect outside human hearing on 96000kHz sample:
* 0.9990234375 = 1 - (pi * 2 * f / 96000)
* f = 14.9208Hz
*
* Confirm this has effect outside human hearing on PDM 16kHz sample:
* 0.9990234375 = 1 - (pi * 2 * f / 16000)
* f = 2.4868Hz
*
*/
const uint8_t dc_filter_shift = 10;
const size_t bytes_per_sample = this->audio_stream_info_.samples_to_bytes(1);
const uint32_t total_samples = this->audio_stream_info_.bytes_to_samples(data.size());
if (total_samples == 0) {
return;
}
int64_t offset_accumulator = 0;
for (uint32_t sample_index = 0; sample_index < total_samples; ++sample_index) {
const uint32_t byte_index = sample_index * bytes_per_sample;
int32_t sample = audio::unpack_audio_sample_to_q31(&data[byte_index], bytes_per_sample);
offset_accumulator += sample;
sample -= this->dc_offset_;
audio::pack_q31_as_audio_sample(sample, &data[byte_index], bytes_per_sample);
int32_t input = audio::unpack_audio_sample_to_q31(&data[byte_index], bytes_per_sample);
int32_t output = input - this->dc_offset_prev_input_ +
(this->dc_offset_prev_output_ - (this->dc_offset_prev_output_ >> dc_filter_shift));
this->dc_offset_prev_input_ = input;
this->dc_offset_prev_output_ = output;
audio::pack_q31_as_audio_sample(output, &data[byte_index], bytes_per_sample);
}
const int32_t new_offset = offset_accumulator / total_samples;
this->dc_offset_ = new_offset / DC_OFFSET_MOVING_AVERAGE_COEFFICIENT_DENOMINATOR +
(DC_OFFSET_MOVING_AVERAGE_COEFFICIENT_DENOMINATOR - 1) * this->dc_offset_ /
DC_OFFSET_MOVING_AVERAGE_COEFFICIENT_DENOMINATOR;
}
size_t I2SAudioMicrophone::read_(uint8_t *buf, size_t len, TickType_t ticks_to_wait) {

3
esphome/components/i2s_audio/microphone/i2s_audio_microphone.h
View File

@@ -82,7 +82,8 @@ class I2SAudioMicrophone : public I2SAudioIn, public microphone::Microphone, pub
bool correct_dc_offset_;
bool locked_driver_{false};
int32_t dc_offset_{0};
int32_t dc_offset_prev_input_{0};
int32_t dc_offset_prev_output_{0};
};
} // namespace i2s_audio