Audio conversion (to wav file) from PDM mic output is having loud noise

[jaydeep.faldu]

Hi,

Hope you are doing good.

I am trying to convert PDM mic output data to audible form(means in .wav file by adding wav header). Below is my setup.

Development kit : ESP32-S3-MINI-1
ESP-IDF version : 4.4.1 (stable release) and 5.0
IDF example : i2s_audio_recorder_sdcard (in IDF 4.4.1) and i2s_recorder (in IDF 5.0)
PDM microphone : SPH0641LM4H-1

I am getting PDM mic's output data over terminal via UART(by printing HEX data). And then converting it to ".wav" manually. But when I listen the ".wav" file then audio is having loud noise also and output is also not proper. Below is my testing code (for IDF - 4.4.1 and i2s_audio_recorder_sdcard example).

Code: Select all

void record_wav()
{
    static uint16_t i2s_readraw_buff[SAMPLE_SIZE];
    uint32_t rec_time=2;
    int flash_wr_size = 0;

    uint32_t flash_rec_time = BYTE_RATE * rec_time;

    printf("Record starting now....flash_rec_time = %d!!!!\n", flash_rec_time);
    // Start recording
    while (flash_wr_size < flash_rec_time) {
        // Read the RAW samples from the microphone
        i2s_read(CONFIG_EXAMPLE_I2S_CH, (char *)i2s_readraw_buff, SAMPLE_SIZE, &bytes_read, 100);

        for (int i=0; i < SAMPLE_SIZE/2; i++)
        {
            printf("%04x", (uint16_t)i2s_readraw_buff[i]);
        }

        flash_wr_size += bytes_read;
        vTaskDelay(50 / portTICK_PERIOD_MS);
    }

    while(1){
    	vTaskDelay(200 / portTICK_PERIOD_MS);
    }

    ESP_LOGI(TAG, "Recording done...... Uninstalling driver.....!");
    ESP_ERROR_CHECK( i2s_driver_uninstall(CONFIG_EXAMPLE_I2S_CH) );
}

void init_microphone(void)
{
    // Set the I2S configuration as PDM and 16bits per sample
    i2s_config_t i2s_config = {
        .mode = I2S_MODE_MASTER | I2S_MODE_RX | I2S_MODE_PDM,
        .sample_rate = CONFIG_EXAMPLE_SAMPLE_RATE,
        .bits_per_sample = I2S_BITS_PER_SAMPLE_16BIT,
        .channel_format = I2S_CHANNEL_FMT_ONLY_LEFT,
	.communication_format = I2S_COMM_FORMAT_STAND_MSB,
        .intr_alloc_flags = ESP_INTR_FLAG_LEVEL1,
        .dma_buf_count = 64,
        .dma_buf_len = 200,
        .use_apll = 0,
    };

    // Set the pinout configuration (set using menuconfig)
    i2s_pin_config_t pin_config = {
        .mck_io_num = I2S_PIN_NO_CHANGE,
        .bck_io_num = I2S_PIN_NO_CHANGE,
        .ws_io_num = CONFIG_EXAMPLE_I2S_CLK_GPIO,
        .data_out_num = I2S_PIN_NO_CHANGE,
        .data_in_num = CONFIG_EXAMPLE_I2S_DATA_GPIO,
    };

    // Call driver installation function before any I2S R/W operation.
    ESP_ERROR_CHECK( i2s_driver_install(CONFIG_EXAMPLE_I2S_CH, &i2s_config, 0, NULL) );
    ESP_ERROR_CHECK( i2s_set_pin(CONFIG_EXAMPLE_I2S_CH, &pin_config) );
//  ESP_ERROR_CHECK(i2s_set_pdm_rx_down_sample(I2S_NUM_0,I2S_PDM_DSR_16S) );
    ESP_ERROR_CHECK( i2s_set_clk(CONFIG_EXAMPLE_I2S_CH, CONFIG_EXAMPLE_SAMPLE_RATE, I2S_BITS_PER_SAMPLE_16BIT, I2S_CHANNEL_MONO) );
}

void lv_tick_task(void *arg)
{
    while(1)
    {
        vTaskDelay((200) / portTICK_PERIOD_MS);
    }
}

void app_main(void)
{
    ESP_LOGI(TAG, "PDM microphone recording Example start");

    // Init the PDM digital microphone
    init_microphone();
    ESP_LOGI(TAG, "Starting recording for %d seconds!", CONFIG_EXAMPLE_REC_TIME);
    // Start Recording
    xTaskCreate(lv_tick_task, "lv_tick_task", 1024, NULL, 1, NULL);
    xTaskCreate(record_wav, "i2s_pdm_rx_task", 4096*5, NULL, 5, NULL);
}

So please suggest me the proper steps if I am doing anything wrong. I doubt on decimation (while printing HEX data on terminal and then converting it to ".wav" file). Do we need to do proper decimation externally for clear audio? I also tried by adding Biquad-LPF using ESP-DSP library but in this case audio file is having loud noise only.

Any kind of help is extremely appreciated.
Thanks in advance.

[ESP-Kevin]

Hi @aydeep.faldu,

I see you printed all the data in the `i2s_read` thread and delayed 50 ms every time. Unfortunately, it will lead data lost, which should be the reason of the heavy noise.

Although there are printing and delays in some i2s examples, they are only used for monitoring whether the data is correct by human. In the real application like recoding, suggest making the interval between `i2s_read` as short as possible. As I2S is a continuous peripheral, large latency in reading or writing may cause data discontinuous.

Therefore, the delays must be removed. As for the printing, it is also not recommended, because it is too slow to catch the I2S read speed. Normally, we suggest to write it directly into a SD card or flash like `i2s_record` example.

One more tip, you can register a `on_recv_q_ovf` callback on v5.0 or provide an event queue handler while installing on v4.4 to monitor if there is data lost. And when the data lost occurs, there is a method to calculate an appropriate dma descriptor number and frame number to prevent that, you can refer to https://docs.espressif.com/projects/esp ... -data-lost.

[jaydeep.faldu]

Hi @ESP-Kevin

Thanks for your response.

Actually I set delay in i2s_read() to prevent watchdog reset. But this delay can be removed and as per the requirement, it don't need to store the data on SD card or flash. That's why I am printing the data.

Meanwhile, as per your suggestions I removed delay from i2s_read()(continuously reading the data from i2s for 2sec or 3sec whatever) and then printing whole data after reading from i2s is completed but noise is still there. (and for now we can assume that data lost is acceptable). For the reference I attached the PDM mic's raw data file and converted .wav file.

So my question is like that do I need to add any low-pass/high-pass filter to remove the noise from the data ? (I tried by adding IIR-BIQUAD lpf using ESp-DSP library, but audio is having noise also). can you please suggest me thr next step to get clear audio without noise ?

Thanks and Regards,
Jaydeep Faldu

[jaydeep.faldu]

Hi @ESP-Kevin,

Hope you are doing good.

Did you get a chance to look into it (as per my last reply) ? Any suggestions or help that I can try on it then it will be appreciated.
Thanks in advance.

Regards,
Jaydeep Faldu