I want to record with multiple scenarios, and multiple recordings, all during a performance. In it's simplest form, have the footswitch record as long as held down, then playback in sync as a loop once the footswitch is released. This sounds pretty straight forward, but believe me it's not! ...well unless you don't care about sync/alignment and tempo...
The key is to keep it simple to use, but have several types of functionality. Here's what I've thought up:
Record/Playback Modes:
 1) Single Record/Playbacks using footswitch sequences, 1 tap/hold rec, release/tap play
 2) Single/Multiple Record/Playbacks as above but start/stop aligned to next beats or bars
 3) Midi Song/Loop triggered Record and play, automatic. i.e. a certain loop records, another plays
 4) Pitch Shift audio samples to match a key change, mostly FX & backing voice

Update: Catastrophic Failure!
 Changing the Recorder Function to Sample Player, and designing a new Sampler. See Below.

 The Chip:
 I've used ISD recorder chips for over a decade now for SFX, microcontroller projects prompts, even the stereo system I built for our boat, but they were pretty straight forward and didn't have an official uP bus of any kind. The audio quality was OK for a 2" speaker in a box, or robot voices, but for any kind of music, even drums, it sucked terribly. The quality reminds me of those old Bucket-Brigade echo/reverb chips. Poor sample rates around 6K/S make aliasing a nightmare for anything other than voice.

 I was trying to figure out how much overhead it would be to record digital audio with CD quality. It's a big task to accomplish! There's some amazing boards, like the one I am using for this projects on-board MIDI Synth, the VS-1053. This chip can encode to MP3/PCM/Ogg and stream via a high-speed serial bus for record and playback. But for sample rates beyond 8k/s, one would need to have an ARM Cortex level processor. Hell, might as will put in a PC-duino!
 Anyway, these limitations left me a bit discouraged, and I sauntered over to the now Novoton's site to see if they had developed any chips with better sound quality. Lo-and-behold the ISD15100!
This chip is a huge jump above the rest of their line. It can record in ADPCM/uLaw and PCM up to 12 bits@32k/s! Totally awsome I thought.

 ISD15100 Issues:
 The actual design level datasheet was extremely hard to get a hold of, the demo-board data sheets, which are an abridged version to use with their developers software were easy to find, but useless. It appears this chip has been designed for automated telephone answering systems, and the dev. software reflects that. Another downer on this chip, as with so many others, it's a LQFP case with .5 mm pin spacing, not exactly human friendly. Digikey has their smaller demo board/breakout but it's $32 CAD and only has the 2 minute chip on it. ( They go up to 16 minutes @ 8khz/4bit ADPCM ) I want the 4 minute which is about $12 CAD, so guess I'll be soldering this puppy.
BTW I'm getting a proper reflow solder station with pre-heat, so hopefully that'll do the job!

 ISD15100's Sample addressing:
Another issue with this chip in my design is the addressing. It is divided into 4096 byte chunks. While this simplifies design and interfacing, it makes addressing a 125 mS step. Now normally, this isn't a problem, but after a stop record, the record doesn't actually stop until the present 4096 chunk's end. This makes back to back looping impossible, unless all the tempos were in multiples of .125 seconds. The bizarre thing is, the "chunk" address is 3 bytes long. It only needs to be 2...I think
 The other problem "chunk addressing" presents is playback in a different (maybe even slightly different) tempo. If the tempo has changed since the record happened, then "fine" adjustments can't be made to a 125 mS address. The master oscillator can be controlled by an external resistor, or source, but if you've ever played with samples in a sequencer, it doesn't take much of a tempo change to result in a noticeable shift in frequency of sample.

 Another, albeit elaborate, method is to slow down/speed up the master osc. to match tempo, then use a pair of NE612's to up-mix into RF frequencies (455kc), filter, then down mix with a fixed oscillator to recover the audio at the original frequency. This method will use lots of hardware so it's not my favorite!

 The last solution would be to start a new recording on every beat (diag. lower right). If tempo speeds up or slows down, then the next playback would cut the last short a bit, or just go quite. The going quiet part isn't a good thing at all, but it may be covered up by the "beat" or even produce some cool FX.
 For the slower tempo (record tempo would be kept) the ISD15100 has a DC voltage ramp-off to zero. This will stop "clicks" at the end of samples, but if the tempo is faster, this would need to be disabled or there'd be a space between each, and unknown delay values. The cryptic datasheet doesn't mention what the ramp speed is, but I imagine it'd be at least 20mS.

Another method, albeit not a solution, is to just stay in tempo. If the sample is long, then that will be pretty intuitive in the same way keeping tempo when playing in a band is, even without percussion.

* Notes:
 The way this works is each action on the footswitch executes the action in that step. The step is then incremented to the next (unless loop to/hold at.) If >8 steps, it loops back to first.
 The "Touch" Record triggers cause a continuous record until it reaches a user set Bar limit, or end of memory, or footswitch is touched/held again.
 The "Hold" Record/Play starts record/Play and stops recording/playing upon release, regardless of beat/bar. Good placement of each type is important as shown in Preset#2
 There can be a Max of 4 different Records in any step sequence, they should be in order 1,2,3,4

The user (me!) can set each of 4 Presets to however it's easy/memorable. A couple of things regarding the difference between "touch" and "hold":
 - Touch must be short enough that it isn't considered a "hold". (User Settable 50mS to 250mS)
 - The action in the present step will begin to execute after the release of footswitch for touch, but after max touch time out for hold.
 - "Loop to" acts on release of touch, or if held, then after time-out of touch as above. Loop Destinations "Touch" will be executed on next beat / bar
 - "Hold at" acts as above, but doesn't execute that Action, just waits until footswitch touched/held. This allows either action to be selected, unlike "loop to".
 - To skip over a record in touch, (to get to next step) just touch and let it record and next step will stop it.

*Possible issues with recording this way might be:
 When should the recording start? At a Bar or at the beat before a bar. I've played around with long echo's (which is  a form of recording really) in a sequencer and noticed that starting a recording right on a Bar isn't always good, especially with vocals. It's hard to notice unless you're thinking about it, but most of the time, the vocals start slightly before a bar, a sort of lead-in I suppose. I've thought about having an offset continuous recording loop going if "record/play" is set up as soon as there is a proper tempo, but this might wear out the flash memory pretty fast! And the flash memory life (of course) isn't mentioned in the datasheet. I guess it could be an option...
Another way would be to use the chips "peak detector" (set low) to detect the start of audio on the input and mark where in the beat it was, then duplicate that in the loop. I think I made that an optional trigger now I'm thinking about it. Still that doesn't really help much unless it is going to be a release to play action.
 I think it basically comes down to the user, me, knowing when to kick that Record in!
 

 MIDI Song/Loop Control:
 This mode is totally exciting to me because I've "sorta" used it before. Last Year a couple of my songs have what I call "word echos" where I sing the last word of a line and FL Studio has "recorded" it into a long delay, which then repeats on every 2nd beat etc. With running EQ FX, this is a super cool sound. With the Audio layout the Playback can be routed to an EQ as well.
Anyway, being able to record selected elements of a song as they're played into the mixer, whether it be vocals, guitar, piano, fart sounds whatever (lol) offers an amazing "live-ness" to the song. Implementing that into this circuit is just a matter of using the pre-ordained MIDI Channel 16's special notes to easily turn on/off record segments, then play them back in the song as a loop or just a one-shot. Because there is to be a MIDI input on the Show-In-A-Box, this can be sequenced live from the computer in a Song Sequence, before creating a MIDI file (time saver!)
 In Loops, it'd be a function of 1 or 2 loop files and to be honest, I'm not really sure how that'd play out (sorry 'bout the pun!)  It'd be done the same way as in a Song though, which could yield some interesting results.
 Recording overdub mixes:
 Because the Recorders Playback can be fed into the STR2 mixer channels, a re-record is possible. Likely there will be some degradation, but it's perfect for layering into a new playback with an extra part in it. If controlled with just the foot pedal, this could become repetitive sounding pretty quick, but with MIDI controlling the instance of playback as well as the EQ of it, it'll sound pretty natural I'd imagine.

 Conclusion Thus Far:

 In any case, I l-o-v-e automation on stage, it's totally liberating to not have to push buttons and switches and just concentrate on singing and playing/tapping. A good example would be the following: A 12 bar blues song in footswitch w/tempo mode:
 Tap up to bar 1 (Tap Pedal not footswitch!) and initiate a MIDI Play of a 12 bar percussion loop. While performing the first verse, 12 bars, hold footswitch to record the basic guitar strumming. At the end of the 12 bars, release the footswitch which will trigger playback on the first beat of the next 12 bars. Now, while still playing guitar you can run off into embellishments up the neck! Once that's done, a tap on the footswitch would stop playback and hold at the playback start position, where a quieter section of the song may be, then touch the footswitch to put the origianl guitar back in.

 If you are reading this and thinking about building something like this, go for it! But really make sure that there is some sort of sequencing going on, even if it's just reading a MIDI Clock coming from a sequencer. From my experimenting I can already see that using record that's in sync is much better than not, us humans just aren't that precise!
 I hope you have enjoyed this "organized description" of the Recorder part of the project. I will be adding updates once I get the chip mounted to a break-out.

May 5th 2014:

Cheers from Sandy in Studio IV

Update:   Because of continuing difficulties with timing, I've decided to release the ISD15104 from the "Sampler/Looper" category, and place it in the "Just Sample Player" Category. This means the ISD board will be used solely for pre-recorded sound effects, where timing isn't super important, and a new Sampler board using an SDRAM, will replace it. Reason:   After I posted this article several problems, besides the 1/8 second issue, popped up. The only solution I could see was to change the osc. frequency and sample rate to close up the gaps. This became easily possible using a DDS, and, with some tempo calculations, seemed to be an answer.   Nope! The alignment shift requires that not only tempo be known, but actions of when/where the loop would be happening so the shift in sample rate can be made BEFORE the recording! This would mean huge overhead on the Main Controller (reading ahead in the MIDI file and interpreting data) and a really confusing layout for commands.   It's been a trip, but it's not worth the effort and time it'd take to perfect. If one can predict "bugs" before writing the software, maybe it's not a good idea to start with.
So I'll post below here with the new "Player" design, and start updating the Sampler page with new info once the project is more mature.
RECORDER