Two I2S Input to .wav

[raviumadi]

Hi,
What's the solution to reading from two stereo external I2S ADC and writing to a single 4-channel .wav file? When I use two I2SStream and two EncodedAudioStream objects, I get so many dropouts in the files. Please help.

[pschatzmann]

You would use 2 I2SStream and one EncodedAudioStream if you want to write to 1 file only. I think the easiest would be just to read one frame of each I2SStream and write it to the EncodedAudioStream.

Assuming you have bits_per_sample=16 it is something like this:

int16_t frame[2];
void loop() {
  i2s_1.read((uint8_t*) frame, sizeof(frame));
  wav.write((uint8_t*) frame, sizeof(frame))
  i2s_2.read((uint8_t*) frame, sizeof(frame));
  wav.write((uint8_t*) frame, sizeof(frame));
}

[raviumadi]

Thanks. Here is what I implemented.
16bit simply does not work. All I get is noise. With 32 bit, however, after 10s of recording (16 or 32bit), I only get 5 seconds.
I have INP441 with ESP32 DevKitc

Update: As I increase the channel number in cfg_out1.channels = 1; to 2, the duration is halved. at 4 channels, it is 2.5s. Is the sampling rate devided between the channels? How do I sort this?

#define FRAME_SIZE 256  // Process 256 samples per loop

void loop() {
    runtime = millis();
    
    int32_t frame[FRAME_SIZE * 2];  // Stores 256 samples from 2 mics (512 values)

    // Read 256 samples from each microphone
    i2s0.readBytes((uint8_t*)&frame[0], FRAME_SIZE * sizeof(int32_t));
    i2s1.readBytes((uint8_t*)&frame[FRAME_SIZE], FRAME_SIZE * sizeof(int32_t));

    // Write the entire frame at once
    wavwriter.write((uint8_t*)frame, sizeof(frame));

    if (runtime > 10000) {
        audioFile.close();
        Serial.println("FILE CLOSED");
        while (1);
    }
}

[pschatzmann]

Your logic is wrong: you need to read and write each frame separatly: with 32 bits it would be

int32_t frame[2];
void loop() {
  i2s0.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame))
  i2s1.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame));

this will give the 4 channels 0left,0right, 1left, 1right... at each loop step in your output file

by using int32_t frame[1]; instead you would get 0left, 1left, 0right, 1right... (also 4 channels, but lined up differently)

[raviumadi]

So, this is not working. All I get is 2.5 seconds of distorted audio. It works alright when I make the wav file two channels and omit the second I2S read-write. I also tried using two buffers, then sorting channel samples in a loop.. can't get it work as intended - 10 seconds long audio with 4 channels :-(

[pschatzmann]

I suggest that you just write out the data to Serial first so that you can check the data in the Serial Plotter

Share your code if it is not too long and messy: maybe I see what you are doing wrong...
Also share the wav file.

[raviumadi]

Ok. Here is the code. And the outputs below. Thanks :-)

1. I am using two INP441 - which is not stereo but I select two channels in config (this mic is a placeholder)
2. When outconfig is 2 channels and only one read-write pair in the loop is selected, the audio comes out fine, 10 seconds and clean.
3. With out config 4 channels and both pairs of read-write, the recording is halved and there is data in all channels - where I expect data only in ch 1 and ch4 (due to L/R selection on INP441)
4. Without FRAME_SIZE the recording is only 2.5 seconds and distorted.

#include <SPI.h>
#include "Arduino.h"
#include "AudioTools.h"
#include "SD.h"

#define SD_CS     5
#define SD_MOSI   23
#define SD_MISO   19
#define SD_SCK    18

// I2S0 - Microphone 1
#define I2S0_BCK   32
#define I2S0_WS    33
#define I2S0_DATA  35

// I2S1 - Microphone 2
#define I2S1_BCK   25
#define I2S1_WS    27
#define I2S1_DATA  26


SPIClass SPISD(HSPI);  // SPI object for SD card

unsigned long runtime;
I2SStream i2s0;
I2SStream i2s1; // Two I2S streams

File audioFile; // one wav file
EncodedAudioStream wavwriter(&audioFile, new WAVEncoder()); // WAV encoder 

void setup() {
    Serial.begin(115200);
    AudioLogger::instance().begin(Serial, AudioLogger::Warning);

    Serial.println("Initializing SD card...");
    SPISD.begin(SD_SCK, SD_MISO, SD_MOSI);
    if (!SD.begin(SD_CS, SPISD)) {
        Serial.println("SD Initialization Failed!");
        return;
    }

    Serial.println("Initializing I2S interfaces...");

    // Configure I2S0 (Microphone 1)
    auto cfg0 = i2s0.defaultConfig(RX_MODE);
    cfg0.i2s_format = I2S_STD_FORMAT;
    cfg0.bits_per_sample = 32;
    cfg0.port_no = 0;
    cfg0.is_master = true;
    cfg0.channels = 2;  // Stereo out but INP441 only provides one channel, the other is zeros
    cfg0.sample_rate = 44100;
    cfg0.pin_ws = I2S0_WS;
    cfg0.pin_bck = I2S0_BCK;
    cfg0.pin_data = I2S0_DATA;
    i2s0.begin(cfg0);

    // Configure I2S1 (Microphone 2)
    auto cfg1 = i2s1.defaultConfig(RX_MODE);
    cfg1.i2s_format = I2S_STD_FORMAT;
    cfg1.bits_per_sample = 32;
    cfg1.port_no = 1;
    cfg1.is_master = true;
    cfg1.channels = 2;  // Stereo out but INP441 only provides one channel, the other is zeros
    cfg1.sample_rate = 44100;
    cfg1.pin_ws = I2S1_WS;
    cfg1.pin_bck = I2S1_BCK;
    cfg1.pin_data = I2S1_DATA;
    // cfg1.pin_mck = 3; // remove this for inp441
    i2s1.begin(cfg1);

    // Configure WAV output for each file
    auto cfg_out = wavwriter.defaultConfig();
    cfg_out.bits_per_sample = 32;
    cfg_out.channels = 4; // chose 2 and omit one of the read-write pair in loop
    cfg_out.sample_rate = 44100;
    wavwriter.begin(cfg_out);
    
    // Open WAV files
    audioFile = SD.open("/audio_4ch.wav", FILE_WRITE);
    if (!audioFile) {
        Serial.println("Failed to open files!");
        return;
    }

    Serial.println("I2S started, Recording started...");
}
#define FRAME_SIZE 512  
int32_t frame[FRAME_SIZE * 2];

void loop() {
  runtime = millis();
  // either of the pair alone works, but with both, the rec is halved and there is data in all channels.
  i2s0.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame));
  i2s1.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame));

  // Stop after 10 seconds
  if (runtime > 10000) {
    audioFile.close();
    Serial.println("FILE CLOSED");
    while (1);
  }
}

when only one pair of read-write is ON with output config 2ch wav

when both pair of read-write is ON with output config 4ch wav

[pschatzmann]

I already told you: the frame size must be 2! You have only 2 input channels and not 512!

[raviumadi]

So, I set 1 ch for each i2s config, set 2 ch for wav out config, with following loop

int32_t frame[2];

void loop() {
  runtime = millis();
  // either of the pair alone works, but with both, the rec is halved and there is data in all channels.
  i2s0.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame));
  i2s1.readBytes((uint8_t*) frame, sizeof(frame));
  wavwriter.write((uint8_t*) frame, sizeof(frame));

  // Stop after 10 seconds
  if (runtime > 10000) {
    audioFile.close();
    Serial.println("FILE CLOSED");
    while (1);
  }
}

and this is what I get,

Only 5 sec distorted audio in both channels.

[pschatzmann]

If you use 1 channel the buffer needs to be int32_t frame[1]; !

[pschatzmann]

If it is distorted you need to do some measurements: the SD drive is just too slow to write that amount of data in the available time!
It is already a challenge to read 32 bit stereo data!

[raviumadi]

It is also lacking the rest of the 5 seconds.
And.... all else being the same, adding FRAME_SIZE and one pair of read-write, i2s config 2 channels, the audio comes full length and clean.

Sorry, I am unable to make sense of this... :-(

[pschatzmann]

If you want to have correct timing you need to calculate the time in terms of samples. For 5 seconds you need to process 5 * 44100 loops (using your current sample rate)

I suggest that you reduce the sample rate down to 8000 and the start to figure out where the limit is...

[raviumadi]

Sampling Rate Update: 8k, 16k and 22.05k work! I get 10s and clean audio. After that, I guess there are dropouts and, therefore, both fewer samples and distortion. For some reason, I can't get INP441 to work at 16 bits.

Please tell me more about the codec method for handling 4-channel data at 192kHz. 16 bits is enough for me and I am hoping the Audiophonics ADC will allow me to use it.

[pschatzmann]

I think AAC would support 4 channels: but it might be quite tricky to get that working as well.
Conversion to 16 bits is easy and since you already handle the buffer yourself you would just convert the int32_t frame[2];
to int16_t frame[2]; by dividing the values with 65538.

[raviumadi]

Thanks. For my application, though, the recording doesn't necessarily have to be in a single file, as long as the samples are in sync in different files. Does that change anything?

[pschatzmann]

I don't think so, the amout of data to be written and the number of write calls stays the same, but maybe it helps if you minimize the number or writes. You can wrap the File in a BufferedStream and write to the buffer instead. The buffer gets flushed automatically when it is full, but you must close it when you are done to flush the remaining data.

const int buffer_size = 1025;
File audioFile; // one wav file
BufferedStream buffer(buffer_size, audioFile);
EncodedAudioStream wavwriter(&buffer, new WAVEncoder()); // WAV encoder 
...
    buffer.end();
    audioFile.close();
...

Try to play with different buffer sizes....

[raviumadi]

By any chance, is there a ring buffer implementation in the library? Because, for my application, typically the the recording is triggered either manually or via sensor, and I would like to record the previous 2 seconds or so upon trigger.
Also, I am considering using PSRAM on ESP32 S3 (8mb) to store the entire file (5-10s) before writing it out to SD at once. Achievable?