4 ms·
> Ambisonic recording is great, though it does have limitations because you're baking in spatial information in a way that will limit editing later Thats not t
by crankylinuxuser 8y ago
> Ambisonic recording is great, though it does have limitations because you're baking in spatial information in a way that will limit editing later
Thats not true as long as
1. You record with enough near field mics
2. Your spatial resolution is fine enough to fully encompass the person's sound
As long as you do and calculate for both (or choose a hardware platform correctly) then audio processing is a simple modification within that bubble.
The problem is tracking the voice as it moves around the room, and correctly mutating the voice with the correct filters, while avoiding modifying data you don't want. My guess, is there will soon be an echo sound test to triangulate the geometry of the walls. With effective calibration and base audiographs, then much more can be done.
Of course, then its 30 seconds before someone feeds CMU sphinx each voice stream and realtime translates it into text.
I've had this workflow going on for about 4 years once I found out the XBox Kinect had 4 near field mics as well as a ir depth sensor and webcam.
If you want a jumpstart on this, install ROS, install HARK https://wp.hark.jp/faq/ https://wp.hark.jp/faq/ and go to a used game store or pawn shop and buy a old style Kinect for $10. You'll need the USB cable so you might need to go to eBay as well.
- anigbrowl 8y agoThat's great supplementary information. I guess I left out a step because I was worrying about the realities of shooting narrative, where people who appear to be sitting opposite (or in some other spatial relationship to) each other on screen may actually be differently oriented, not not even in the same location at the same time, through the magic of editing. My early experiments with ambisonic recording indicated that if sound sources weren't positioned consistently during shooting the editing and mixing turned out to be nightmarish, so I stuck with using surround recording to get nice room tones and then prioritizing the mono recordings of the actors' performances for all but the simplest scenes. Your workflow sounds very interesting and at this low cost I definitely want to give it a whirl, thank you!
- crankylinuxuser 8y agoGladly! One thing to keep in mind, is that when you plug in the Kinect, you can get the ir and webcam data trivially. However, when you load up the appropriate also commands to see the audio device, it will be markedly not present. You need to load the appropriate firmware with this tool ( https://manpages.debian.org/stretch/kinect-audio-setup/kinect_upload_fw.8.en.html https://manpages.debian.org/stretch/kinect-audio-setup/kinec... ). Once you load the firmware, you'll see the inputs and outputs as you should. I forget the exact procedure, but I know you have to download the driver and strip the firmware.
- ganzuul 8y agoWith deconvolution of an impulse response sweep you can get the spatial data in the form of an echo. You should get better data with a sweep than with a single pulse due to the general uncertainty principle.