I am resurrecting this thread to report that at last I have had a chance to use the Zoom M3 with its 32 bit float stereo sound in a real world scenario. I attended a local festival which included market stalls, stage performances, food avenue, and fire sculpture, which began in mid afternoon and ended at 9pm. I used a DJI Pocket 3 camera whose audio capabilities are pretty good for a tiny camera, but I attached with some difficulty the Zoom M3 in order to be able to forget about level setting. Levels varied considerably, from scenes at some distance in the festival field, to standing right by the stage during live performances including a group of three drummers.
I realise that this is hardly a stealth device, but for occassions when stealth is not required, it's very light and simple to use. In essence, it has a power button, a record toggle button, and that's about it. Not even a display.
I was generally pleased with the sound quality, and the stereo image is excellent (and is recorded in both A/B and M/S formats so you can play with the width afterwards). I was careful about how it was mounted to the video rig but I experienced no handling noise issues.
Apart from the fiddly task of lining up the audio and video when editing, the biggest problem was dynamic range. Most video that you see on YouTube has audio which has been captured via automatic gain control, and viewers would be quite disconcerted to hear a video with full dynamic range. And let's face it, a substantial majority of viewers are using a phone where dynamic range means a lot of stuff being inaudible. I decided that I would normalise the audio at the outset, clip by clip, but perhaps this was a mistake. I then struggled to bring down the level of the performances, and bring up the level of the quiet bits, because the editing software I used didn't have enough control range for the task. I then started over from scratch and used audio which I had pre-levelled using Reaper, although it was still a struggle. I used a limiter VST to help matters, but I was careful that it was set only to catch transient peaks.
So you might wonder, what was the point of using 32 bit float audio and then squashing the dynamic range like that? Well, the chief benefit was knowing that clipping at the outset would be very unlikely even close to the stage PA speakers. At one point there is some clipping audible, but that could be heard with my own ears when recording, in other words, the canned music they used was clipped at the outset just for one item. And I had a complete choice over final levels in the quietness of my 'studio' back home.
However, normalising 32 bit float audio from a live recording means that the most trivial and instantaneous peak may dictate the overall level of the result. For instance, someone clapping enthusiastically near to the mic might cause a higher transient level than the performance audio. In non 32 bit float recordings, if that got clipped it wouldn't matter much, because the transient was too short for the clip to be heard. So the use of a limiter might actually be almost required in video post production, depending on the nature of the material. But at least there's a choice, rather than being stuck with the wrong level at the outset.
Here's a link to the edited video - which is rather long but it has chapters which could enable you to skip to the drummers, where the audio is perhaps the most interesting.
https://youtu.be/KCT_izLIJqQ
