Thank you! That's very kind of you to say and I really appreciate it.

I agree 100%. Great work!

That's really useful to know - I'm going to try increasing the number of samples I take to get a larger number for the averaging. Also need to work on the power supply to get the noise down even more.

Thank you - it's really hard to describe any of this stuff in just words.

Glad it was helpful!

Coming soon!

What about the specs? Don't they mention it?

Glad I could help!

Should be possible - for voice you probably don't need to be sampling at 40KHz, you'll get away with 10KHz pretty easily. If you want to get fancy you could take a look at some very low bitrate codecs. Even without that you should be able to do full duplex over WiFi. You may want to look at using UDP for streaming the data as it's a bit more lightweight. I'm going to be investigating my noise issues some more to see how far I can get and I've also got an I2S microphone break out board which will let me bypass the ADC completely. For your project there's also an I2S amplifier breakout board on Adafruit that might be worth looking at.

please, post link if you have chance. thanks sir

Yeah, a good clean power supply is essential for anything audio. I2S microphones seem to be a bit better, but I've still seen some noise issues.

@@atomic14 could that be the reason why the ESP-CAM has noise in the images too? The power supply noise should also affect the camera, shouldn't it? Interesting...

@@atomic14 I used two mics to do SSL (sound source localization using the FFT) let me know if you are interested and I can share with you the details/code

@@raguaviva That sounds really interesting! I'd definitely be interested. We've got a discord group going for the channel if you'd like to jump into that: discord.gg/DJ8EQRXU

@@raguaviva could be yes, depends on what kind of noise you are seeing.

That’s a good suggestion.

What were you thinking of doing?

I'm glad you enjoyed it - I had a lot of fun making it!

@@atomic14 It's good to hear 😃 I'm waiting for another video. Btw, is there a place where I can learn about how to use the ESP32's i2s with the arduino IDE?

@@avieleliyahu6199 You should be able to use all the sample code in the Arduino IDE without any problem. Just copy the contents from main.cpp into the sketch file and then create new tabs for the extra files (WiFiCredentials.h, I2SSampler.h, I2SSampler.cpp) and copy the contents of each file in. You create a new tab using the little down arrow on the right - imgur.com/a/nYwfGwS

No problem. The ADC gets a really bad rap, but I don't think it's that bad.

Your guitar playing is amazing!

This is why I love the viewers of this channel. This is a really informative and useful comment.

I use Audacity - www.audacityteam.org/ - it has an option for importing raw audio data - you just need to tell it the sample format, number of channels and sample rate.

Thank you! You too!

I think the best solution is really putting in a good clean power supply for the microphones. I did find that using the 5v supply with a good LC filter followed by an LDO would give a very clean 3v3 supply for the microphones.

Hi Pieter Snyman Am also trying to work with esp32 s3, How to work on it, with using i2s on board ADC?

@@saranyaasuresh5710 in the end I used a digital microphone instead (that does the AD conversion) linked to I2S on the ESP32 S3. The S3 does not have the capability to use its internal ADC with I2S. As far as I understand this is preferable because the ESP32 ADC is not very good.

Hi Mike, it looks like it should be possible - there's a bunch of information here that might be useful - github.com/espressif/esp-idf/pull/1991

I would use an I2S microphone for input in preference to the internal ADCs as the noise seems to be much lower - (ukposts.info/have/v-deo/a5dooGuZpG2Xu2w.html). For transmitting the audio data, depending on how much bandwidth you have then you can just send the raw audio samples to another device using TCP (or HTTP).

@@atomic14 Around 4:15 of this video you mention picking up the audio on your desktop. Could you explain that setup a little? I'm having trouble getting the I2S audio data to transmit on WiFi. My goal is to have one ESP32 with a digital mic communicate wirelessly to a second ESP, and then play the audio out the I2S amp on that one. I had an analog mic setup working this way with the 8266 version of this card (NodeMCU), but can't seem to do the same with the ESP32 version. Thank you for your excellent videos!!

It's worth a try, I found that that I was getting quite a low volume on the voice but it's definitely worth another look. I'm going to do a follow-up video in the next few days as I have some I2S microphones that I want to try out so I'll revisit the MAX9814 at the same time.

@@atomic14 The 60db with a pi and soundcard is very noisy its 40db or 50db with the Max9814. The timing on the A/R is also really too fast for voice the 100nf cap prob should be 680nf or 1uf.

For audio streaming you can take a look at my walkie-talkie video. I’ve got video streaming coming soon!

You're welcome!

I need to do a proper followup video to this series - I've got a really good collection of mics and pre-amps to use. It's definitely in the schedule. I'm also quite keen on actually quantifying how bad the ADC actually is. Separate the performance of the ADC from the mics and see if it really is as bad as everyone says.

@@atomic14 That would be useful. I've read that it is only capable of: 7 Bits ENOB out of the box, closer to 10 Bits with the recommended capacitor and 64 sample averaging as described in the Espressif docs 6 kHz signal bandwidth (presumably the 3 dB cut-off 150 kHz samplerate Non-linear resoonse and unable to read below about 0.15 V Reading 254 samples continuosly and then skipping 2 samples whilst it does something internally Or reading in "continuous mode" which Espressif recommends against because it drops the quality substantially. Also, apparently, once every 17 seconds there is a window when there is much less noise for a second or so; so any quality measurements will be very different depending on whether you happen to hit that window! I'd really like to see them improve the ADC in future ESP32 versions.

That's interesting, the output should not be able to go negative at all provided you have the GND connected to 0v and VCC to 3.3V. It should be biased at 1.25V on the output and go 2Vpp (so 0.25 to 2.25 volts) - cdn-learn.adafruit.com/downloads/pdf/adafruit-agc-electret-microphone-amplifier-max9814.pdf

@@atomic14 I made a noob mistake. I connected the MAX9814 to my scope and had AC coupling enabled. Switching to DC coupling I see the signal hovering around 1.28V (avg) which is inline with the spec. Thank you again for a great video!

Take a look at this - ukposts.info/have/v-deo/g5F7aJhqf4Bkr3U.html and ukposts.info/have/v-deo/e6d6i6Z5qJdez4E.html I think you are talking about something called a spectrogram.

Apologies for the delay in replying - yes, the DMA and I2S can be used directly with the internal ADC - it doesn't have to be an audio signal.

Definitely possible - you can run input and output simultaneously. In theory you can use a single I2S peripheral for both input and output - though in practice this can be a bit difficult depending on the devices you have. In my Alexa project I'm actually running both an I2S microphone and and I2S amplifier at the same time using both the I2S peripherals. When I don't need any output I just play silence and when I don't need any input I just ignore the samples coming in. Have a look at ukposts.info/have/v-deo/eaRphHqIap1op4U.html and ukposts.info/have/v-deo/a5dooGuZpG2Xu2w.html - you can also have a look at the code for the Alexa project - ukposts.info/have/v-deo/foplla-wi45-k6c.html

On the task function definition, you'll see that it's defined as void task_function(void *param) - the value of the param is what you pass into xTaskCreatePinnedToCore. So this allows you to pass data to the task you are creating. In my case, I'm just passing in a pointer to the current instance.

@@atomic14 Thanks. Makes more sense after downloading the source files and seeing the full thing. I'm used to using plain C as much as possible and didn't realise it was inside a method of a class.

Hi, I am also facing similar issue. Did you find any solution? We can get in touch and discuss. Please ping back so that I can share my email Id with you.

Also please make a same video for that too.

I actually have an I2S MEMS Microphone (www.mouser.com/datasheet/2/218/ph0645lm4h-b-datasheet-rev-c-1525723.pdf) on an Adafruit breakout board on order right now! Soon as it arrives I'll do a video to compare the performance.

@@atomic14 extremely excited to see the comparison bcoz yesterday i see that inside a samsung earbuds they used the INMP441 microphone 😎

@@Dharm1999 Will do!

There should be a semi-colon there. Good spot.

There's a maximum size that you can specify for each DMA buffer (I think from memory it's 1024 bytes) so you can use multiple DMA buffers if you want something larger.

@@atomic14 That explains it! Thanks for responding so fast. I didn't know there was a limit on buffer length. Is there a limit on number of DMA buffers (memory permitting)?

possible, but looks pretty complicated - esp32.com/viewtopic.php?t=17547

You'd need something to convert the signal to I2S - it depends on what format the data is coming out in.

I'm afraid that's a bit more complicated than what I've done here. This might help www.reddit.com/r/esp32/comments/ah9c2t/has_anyone_played_sound_from_an_esp32_to_a/ but it is for ESP32 not Adruino.

Hi Dominik, I was thinking of doing exactly that for my next project - I've had a few people ask about writing audio to an SD Card and thought it might be quite good. On the DMA buffer size and count - there's a maximum size for each buffer of 1024 so if you want more space then you need multiple buffers. The amount of space you allocate for the DMA buffers is a bit of a trade-off. More space means the CPU has to service the I2S device less frequently. But comes at the cost of increased latency. You are correct to pick up on the fact that I also have the double buffers where I'm storing the audio data before it gets transmitted over the HTTP request. My thinking around this was as follows: I wanted to send a reasonable amount of data in one go so that I wasn't trying to make too many HTTP requests in rapid succession. So I needed to buffer up the audio data before sending it. I picked quite a large amount for the DMA buffers to reduce the load on the CPU to as small as possible to make sure the data would be transmitted successfully. It's likely that I could get away with a much smaller total DMA size as there's not really much CPU work going on but I wasn't too concerned about latency. The size of my local buffers where I hold the audio data before sending over HTTP was down to a tradeoff between how rapidly I wanted to send the HTTP requests and how much free RAM the device has for holding the audio data. Potentially, you could increase the total DMA buffer size to the point where you have enough time to send the audio data without double buffering. This is the approach I've taken in my Walkie-Talkie code where accumulate the samples into one buffer before sending them out and rely on the DMA buffers to hold the samples that are coming in. In my Walkie-Talkie project, I've used 4 DMA buffers of 64 bytes which gives about 16ms of audio data when sampling at 16KHz - I'm actually thinking of increasing the size of the DMA buffers to give a bit more time. There's quite a bit of guesswork and assumptions on my side around a lot of how this works. Hope that helps!

@@atomic14 Wow, thank you very much for this fast and explaining anser! The first answer which helped me A LOT was the fakt that the max size of each each DMA Buffer is 1024. I played around with the buffer size but then i got a "paniced" error from my ESP32. Now i know why. However, when i specify the Buffer count and size, it gets allocated at one block in the RAM? Or how does it work? To sad that there is not a good documentaion from espressive. :( Or is there any i didn't found? ). I think it is task controlled. So every time one DMA buffer is full, the task get notified and can read the bytes into a local buffer!? Sad to say, but I'am also not familiar with the FreeRTOS and Task things. I know about multithreading in Java... like race conditions... deathlock situations an so on. In that past i had build an Java Project which was connected to over 3000 gameservers simultainisly with more than 50000 Threads and and including a selfe written HTTP Server to provide an interface for the users. Well... some time went along since I did this... However... I understand now why you use a "local" buffer. I faced some problems while writing 44100hz Audio Data to the SD card. I made some mistakes. Frist Problem was, that i opened the File to write every time when the buffer was full. Second problem was that i wrote every bye as a single byte. The "overhead" from the SD library was way the high. Now i changed the code too write the samples from the local buffer to the file in bigger chunks with the File::file.write(bytesToWrite, numberOfBytesToWrite) funktion. That was the key to get rid of loosing samples. But overall, i didnt really understand the DAM buffers in detail. I would apreciate if you would share your knowlage about it in detail! :) Another think i didn't really get is the const uint32_t usStackDepth from the xTaskCreatePinnedToCore funktion. Seems like to be bit of trail and error, to not run out of stack!? But is has nothing to do with the DMA buffer oder the "local" Buffer, right? THanks a LOT!

@@dominikwinkler2718 No problem - happy I can help! There are some docs here that explain the workings of the DMA system - section 6 is quite interesting www.espressif.com/sites/default/files/documentation/esp32_technical_reference_manual_en.pdf It's basically created a linked list in memory. So each buffer is allocated separately. I think under the hood it's all interrupt driven but with the I2S driver you can give it a handle to a queue and get notified via that when things have happened (a buffer has been filled/emptied etc...). On the stack size - I've found that's very much trial and error as you really don't have any way of really knowing how much stack space you're going to need. I've ended up with the classic programmer approach of using powers of two - 1024, 2048, 4096 until it stops crashing :) As you say though, it's completely independent of the DMA buffer size and only relates to stack usage - so local variables and function calls. You can easily accidentally blow the stack by doing a large local array variable. I must admit, I'm still learning about the embedded world. There are some things I still find very odd in the embedded world. I'm finding Arduino libraries particularly strange - they really love their global variables.

​@@atomic14 So nice to getting i touch with you! Thanks! I didn't found THIS reference manual you linked so far! That, helps me again a lot. Sorry... as i mentioned... I'am quite new to this ESP stuff! Another thing: Shouldn't we call call i2s_stop(I2S_NUM_0) befor we call i2s_read(....., ...) ? and then - after reading from the DMA buffer - call i2s_start(I2S_NUM_0) to prevent reading interrupted data? About the stack thing.... i dont worry at the moment... but i was right that it has nothing to do with the DMA buffer, thats good so far! THank you! However: You might be new to the embedded world... but you are definitly NOT new to c++ programming. You are using classes... sub- or child classes ... using pointers where ever you can to safe time and memory. Thats great if you are a good c++ coder.... but to be honest, for a youtube tutorial, it might be a little bit of overkill for some viewers. It took me some time get back into all that c++ class stuff. For an easy tutorial video for the "mainstream" .... for most of the Arduino "Makers"... it might be hard to understand. I think that is the reason why "Arduino" lover their global variables so much. They want to keep it as simple as possible to make it possible fore everyone to get started with coding and electronics. During my informatics studies i faced a LOT of fellow students who didn't even unterstandstand what a pointer is. Things like * or & ...... ooooohhhhh they got feared :D . What is an memory address ? Why can i cast any number to a void* and just put data at this point of the memory (if exists ^^) .... and why can i user [x] brackets on a pointer and just write to another adress even if i never declared an array ^^ ? An What the heck is malloc()? In Java or any other managed language I don't even care.... i just declare an var... or list... and the garbage with take care about it. What i try to say is: for many people (including me) its hard to understand all this deep C++ or asm stuff. Well, in the past i reengineered .exe files with ollydbg and wrote sometimes memory cheats for games. So when you wrote your frist on the fly memory hack/cheat to make a JUMP to a part of the free heap and placed your own peace of code there in asm or placed some NOP's to get rid of damage if you are tackled .... then you know what I'am talking about. but... thats totally nerd shit! ^^ . But i did it about 20 years ago. If iam allowed to give you a hint: Try to make your brilliant video tutorials even better if you write the code you present as EASY as possible! This might give you even more subscribes! Especcially on this i2s topic.... because there aren't much good tutorials. Lets keep in contact :) .... I'am looking forward to your next tutorials! :)

@@dominikwinkler2718 Thanks Domonik - I think you are right. I sometimes forget that I have a lot more experience of C++ than a lot of people. Once you've learned something it's sometimes hard to remember what it was like when you didn't know it. The i2s_read takes care of any synchronisation - I believe it will only read from completed buffers. Though I'm not sure what happens if the i2s peripheral wraps around the DMA buffers. My C++ is actually very out of date now. The last time I worked in C++ was probably about 15 years ago! But I am thinking of doing a series of "basic" C++ videos.

They are probably good enough for your use case. The quality won't be amazing, but it's pretty hard to get good quality anyway.

Hi Rafael, I should add some commends to the code base explain what is happening - it is a bit weird. When reading from the ADC you get a 16bit value from the I2S interface. The top 4 bits of that are the ADC channel that is being read from. So to get the ADC sample we need to remove those top 4 bits This is what this line of code does: sample = sample & 0xfff That give us 12 bits with sample information in it. The values run from 0 to 4095 and are inverted - so 0 is the maximum value and 4095 is the minimum value. So the next line of code translates that into 2048 for the maximum value and -2047 for the minimum value: sample = 2048 - sample The * 15 is just amplifies the value up so we can hear it clearly - in theory I think we could just shift left by 4 bits to get the full 16 bit range. I think the code should just work on the ESP-IDF - you will need to remove any reference to Arduino.h - I think the only thing it is bringing in are types like uint16_t which you can get from #include If you have any problems them it's probably easier to open an issue on the GitHub repo as it's a bit easier to share code there. Good luck with your project! Chris

Hi there, it's not something that I have looked at yet - you might be able to find some inspiration here - github.com/espressif/esp-adf/tree/master/examples/player/pipeline_bt_source

@@atomic14 thank you for helping

I have a simple animation library I've written to help me show waveforms and run simple processing. Unfortunately it's not really suitable for sharing as the code is very messy.

@@atomic14 I understand. Thank you so much for replying !

I think you can - provided the esp8256 supports reading samples from the ADC using the I2S peripheral. I'm just about to post a new video using some microphones that support I2S directly which may be easier.

Next video is up - ukposts.info/have/v-deo/a5dooGuZpG2Xu2w.html

I got and ASP32, the server is ok, Arduino script is compiling normally, the raw file is recorded to the server, but no voice while converting RAW to WAVE

Couple of things you can check - when converting the raw data make sure to use Signed 16 bit for the samples and 1 channel. You can also check the contents of the file to make sure it is not all 0's. I also found when debugging that logging out the min and max values of the buffer very helpful to see if any data was coming from the microphone.

@@atomic14 thank you for your quick reply, i was reviewing the code, I found that you are reading the GPIO_NUM_25 and not GPIO_NUM_35 as in the video. is my note is correct ? or I have some miss understanding?