Skip to content

Latest commit

 

History

History
220 lines (153 loc) · 9 KB

README.md

File metadata and controls

220 lines (153 loc) · 9 KB

PiCam audio spectrum timline

screenshot

Creates a HTML/JS timeline of video captured by a Raspberry Pi camera, along with a corresponding audio spectrum.

video demonstration

It works in combination with the RPi Cam Web Interface. Using the start_vid.sh and env_vid.sh macros, it gets "informed" about video start end end times, which are just stored in the events.txt file.

The spectrum is generated using Python and Scipy. Audio is stored as FLAC to preserve the very high frequencies (above 15kHz). I'm using for my bird box in which the (blue tit) chicks make very high-pitched sounds, which (I hope) will be interesting to analyse.

Usage

Click and drag to pan.

Mouse wheel to zoom in and out.

Click a videos' rectangle (above the line) to watch the video.

Click the spectrum to play the audio from the clicked timestamp.

When you try to drag over a video, it will play from there and not pan the view. Try doing it in the "safe area" directly over the timeline.

This is absolutely not made for mobile devices, I don't know how it would behave there.

Installation

Of course, you need a microphone attached to your Pi. I'm using two Adafruit SPH0645 breakout boards for stereo sound.

Install Raspberry Pi OS lite & do updates. I'm using Bullseye on a Pi zero w and a 32GB microSD card.

Connect your camera and microphone(s).

Install some needed packages:

sudo apt install git python3-numpy python3-scipy python3-pil python3-soundfile ffmpeg lame

Install RPi Cam Web Interface (long form found on the wiki).

# Enable Interface Options > legacy camera
sudo raspi-config 

git clone https://github.com/silvanmelchior/RPi_Cam_Web_Interface.git
cd RPi_Cam_Web_Interface
./install.sh

install.sh settings: I'm using nginx as webserver, it should work with the others too. Use cam as subfolder, as that's what the timeline expects.

Set up audio recording

sudo pip3 install --upgrade adafruit-python-shell
wget https://raw.githubusercontent.com/adafruit/Raspberry-Pi-Installer-Scripts/master/i2smic.py
sudo python3 i2smic.py

Answer y for the "auto load at boot" question and accept the reboot.

After reboot, check if the microphone is available:

arecord -l

You should see the I²S microphone as sndrpii2scard

Set permissions to allow recording. Add root and www-data to the audio group:

sudo usermod -aG audio root
sudo usermod -aG audio www-data

Now, let's tell the web interface to record an audio file while video recording.

sudo nano /var/www/cam/macros/start_vid.sh

Add the following content:

#!/bin/bash
echo -n "`date +%s` " >> /var/www/events.txt
AUDIOFILE=$(echo "$1" | cut -c 1-42)
AUDIOFILE="$AUDIOFILE.wav.gz"
arecord -D dsnoop:0,0 -c2 -r 48000 -f S32_LE -t wav -V stereo | gzip -1 > $AUDIOFILE

CTRL+X, y, enter to save

Add this to /var/www/cam/macros/end_vid.sh

#!/bin/bash
pkill arecord
echo -n "`date +%s` `echo "$1" | cut -c 20-42`" >> /var/www/events.txt
echo "" >> /var/www/events.txt

Make them executable

sudo chmod +x /var/www/cam/macros/start_vid.sh
sudo chmod +x /var/www/cam/macros/end_vid.sh

Change the boxing command to decompress the wav, encode it to mp3 and include it in the final mp4 video.

sudo nano /etc/raspimjpeg

Move down to the line starting with MP4Box_cmd

Delete (or comment out) the entire long line and replace it with this even longer one: (It has to be in one single line!)

MP4Box_cmd (set -e;FPS=%i;TNAME='%s';FNAME='%s';TNAME='%s';AUDIOFILE=$(echo "$TNAME" | cut -c 1-42);LOGS="$TNAME.log";rm -f "$FNAME";gzip -d -c "$AUDIOFILE.wav.gz"|lame - "$AUDIOFILE.mp3";if MP4Box -fps $FPS -add "$TNAME" -add "$AUDIOFILE.mp3" -delay 2=-100 "$FNAME" > "$LOGS" 2>&1;then touch -r "$TNAME" "$FNAME"; rm "$TNAME" "$LOGS" "$AUDIOFILE.mp3";else mv "$TNAME" "$TNAME.bad";fi;) &

The command was taken from this issue on RPi_Cam_Web_Interface and changed a bit. What is does:

  • Get the command line arguments
  • unzip the .wav.gz file and pipe it into lame for mp3 encoding.
  • run MP4Box to combine the .h264 video recording and the mp3 audio.
  • if there was any error in MP4Box, keep the log file and rename the h264 file so it won't try again.

Open /var/www/cam/config.php and add the following two checks to the deleteFile function:

if (file_exists(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.h264.wav.gz")) {
    $size += filesize_n(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.wav.gz");
    if ($del) unlink(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.wav.gz");
}
if (file_exists(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.flac")) {
    $size += filesize_n(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.flac");
    if ($del) unlink(LBASE_DIR . '/' . MEDIA_PATH . "/$rFile.flac");
}

This will also delete the .wav.gz and .flac files when you delete a video using the web interface.

You can now try to record a video using the web interface. When playing it back, it should have sound in it.

If you encounter any errors here, try running the web interface with ~/RPi_Cam_Web_Interface/debug.sh and post an issue.

Set up the timeline

Install php gd:

sudo apt install php-gd

Download the index.html, imagegen.php, hourglass.png and js files and put them into /var/www/

Download jquery and save it as /var/www/jquery.js

Create a directory called spectrum inside /var/www/ and put the spectrum.py, spectrum_queue.sh and turbo.png (colormap) files there.

Record some videos. They should show up as individual lines in /var/www/events.txt

Example events.txt content:

1674394076 1674394107 vi_0014_20230122_142756
1674397385 1674397435 vi_0016_20230122_152305
1674662514 1674662590 vi_0017_20230125_170154

Now, create the spectrograms for them:

/var/www/cam/spectrum/spectrum_queue.sh

Depending on the Pi generation you're using and the video duration, this may take some time.

I don't want to include the spectrum_queue.sh call in the boxing command because it would (especially on my pi zero w) cause 100% CPU usage for some time and therefore interfere with video recordings that would be captured right after the first one.

What are those image tiles?

Read explaination

After recording, the spectrum images are generated by Python + Scipy from the audio track. There are two subfolders per video under /var/www/spectrum/.

folders

Depending on the length of the recording there are several images in the subfolder.

imagefiles

Each image is 257 pixels high, that's how many frequency bins are calculated by Scipy's frequency analysis. Each frame corresponds to a sound duration of 10 seconds and is 714 pixels wide. In the second folder with _s at the end all images are stored with the same height but a width of only 30 px, this saves loading the large images when zoomed out far.

But these images are not transferred to the user's browser. Rather, the width of the screen is split into 300 pixel wide tiles, and for each of these tiles the timeline requests an image from the imagegen.php script.

This gets the position of the tile on the timeline, the zoom level (which time is in its 300px range) and the height of the output image. It then checks if a video with spectrum is in the current tile, loads the matching spectrum images from /var/www/spectrum/, scales and crops them and sends this as a JPG image to the browser. While the timeline waits for the images from PHP, it displays the hourglass.

Such a tile as returned by PHP:

imagegen_tile

Audio data flow

Audio is directly compressed by gzip (flac encoding doesn't like the S32_LE format and converting would be too expensive on recording time.)

arecord | gzip -1 > audio.wav.gz

On boxing time:

gzip -c -d audio.wav.gz > lame | audio.mp3

before spectrum generation:

gzip -d -c audio.wav.gz > audio.wav

after spectrum generation:

ffmpeg -i audio.wav -filter_complex '[0:a]channelsplit=channel_layout=stereo:channels=FL[left]' -map '[left]' audio.flac

Only one channel of the stereo recording is retained, but it is losslessly encoded for further analysis.

TODO

  • When deleting a video using the web interface, it doesn't delete the line from events.txt
  • When playing audio: Shift view when playhead leaves visible area
  • Improve design of the control buttons, create a better menu for them. With direct link to RPi Cam Web Interface

Conclusion

I have programmed this timeline for and directly on my RPi Zero watching a bird nesting box. If there is interest, I can create an automatic, convenient installation script. Feel free to give it a star or create an issue if you are interested.