Advanced sound recorders on the Pocket PC

Up until late 2002, there were no real MP3 encoders for the Pocket PC. This meant anyone that wanted to record on his or her PDA needed to resort to the built-in, either very wasteful, non-compressing WAV coders or (on the PPC2k2 platform; in PPC2k and earlier, there was another ultra-low-speed codec, the Mobile Voice, with even worse voice quality and smaller bandwidth usage) the GSM codec, which, being a heavy-compression vocoder, produced just intelligible results and was completely useless for recording for example meetings.

Then (late 2002) came NoteM (http://www.zavorine.net/wince/mp3.htm), the first MP3 recorder, first with 11 kHz sampling frequency with 16 kbps only. Shortly thereafter, early 2003, a 44 kHz, high-quality codec with 56 and 128 kbps speeds was added.

The competition was quite late. Even as of late 2003, the two other competing recorder applications, Resco Audio Recorder (http://www.resco-net.com/AudioRec.asp) (Resco) and VITO Technology's SoundExplorer (http://vitotechnology.com/products/product_27.html) (VITO) was unable to record at 44 kHz - its users needed to resort to 11 kHz. However, even then, they both contained VAS (Voice Activation System), which can be, at cases, very handy. (You may want to read my old article on them, dated back in Aug. 2003, mostly on their WM2003-compliance.)

However, now, early 2005, both Resco and VITO are able to record at 44 kHz. Furthermore, Resco has some other coders, so a real, comparative test became necessary.

Are current PDA's able to record in high-quality at all?!

Well... yes and no. Depends. Unfortunately, very few of the recent PPC PDA's (the Toshiba e series, the Dell x50v, the iPAQ 5450/5550/hx4700) have a direct mike input ring in their headphones jack. Furthermore, it's mono only. There're no PDA's with stereo sound inputs at all. Furthermore, it's impossible to come up with a CF/SD card that would have direct sound feeding via the card bus to the inner hardware - the current card buses just don't support this kind of functionality. Incidentally, this is why there're no TV/radio/GSM CF/SD cards that could route their sound through the host PDA.

Note the word 'direct'. The, up to now, only external digitizer card, Core-Audio's PDAudio-CF (http://www.core-sound.com/pdaudio-cf.html) card digitizes the audio, and only after that sends it over the CF bus. This is certainly possible because it's simple data communication. However, (it seems) it's impossible to send over any non-pre-digitized (that is, analog) signal over the CF bus - it's just not have a direct, analog PIN (or PIN's) to communicate analogue data with the outside world. If it had, someone would already have come up with a CF-based GSM card/radio/TV receiver/etc card that is also able to route its sound through the PDA. See for example http://www.pocketpcthoughts.com/forums/viewtopic.php?t=36151 for more info on this subject.

And, up to now, there're no quality (meaning: not just GSM-quality), let alone stereo/multi-channel, BT input profiles for PDA's, so you can't use BT either to input high-quality sound into your PDA.

So, is it worth at all to use PDA's for quality sound recording, using a (vo)coder considerably better than the built-in GSM? The answer is yes. Much as the sound quality of current PDA's is severely limited compared to even basic Hi-Fi equipment, they're certainly better (and, compared to digital ones, much cheaper) than current, dedicated, digital (and, for that matter, analogue) dictaphones.

I've compared the frequency response, the noise level etc. of a lot of PDA models and found out that most current devices have tolerable frequency response, which makes using at least 22 kHz sampling frequencies and decent, 16-bit encoders highly recommended. Let's kiss GSM encoders good-bye - fortunately, recent PDA hardware has quite good microphones and sound input.

Do you want to know how exactly some previous-generation PDA's sound? Browse these example recordings. Some explanation: there're two subdirectories with 4 child directories each (for the iPAQ 3630, the iPAQ 2210, the iPAQ 5450 and the Asus 620BT) in the ZIP files. The two main subdirectories separate the sample recordings of the four above-mentioned devices based on how loud the test music was. The 'quiet' subdirectory (it's separated into two ZIP files so that I can upload it to the server) contains recordings I've made with quiet sound source. The point in this test was to find out how the PDA's behave in situations like this: what they record, is the recorded sound hearable at all, doesn't it get drown in the sea of noise etc. The 'loud' subdirectory (two loud*.zip files), on the other hand, used pretty loud sound to find out how much you can 'stress' the PDA hardware: will it record loud sounds with heavy distortion or not. Please note that, as the Asus 620BT has a built-in gain control, I've tested it with it set to minimal and maximal levels.

The test music was 'Tervetuloa (Intro)' from the album 'Sähkönsinistä Sinfoniaa' from the Finnish hard rock band Kilpi (http://www.kilpi.com/ ). I've chosen this song because it has both strong basses and trembles.

I've recorded the sound clips at the same time with the four PDA's, paying special attention to their microphone's seeing just the center of the loudspeaker playing the test music (as was the case with the other tests detailed in this review). You can see the test setup here.

What about the hardware compatibility?

Unfortunately, WinCE/PPC versions up to WinCE3.0/PPC2k only allowed for using 11 kHz sampling frequencies. This is the reason why you shouldn't try to get a decent MP3 recorder for your, say, Jornada 680 or Casio E-125. Unfortunately, you can't surpass the op. system limitations in these devices, even if their CPU's were able to run an efficient MP3 encoder algorithm like that of NoteM. PPC's starting with PPC2k2, however, can use as high sampling frequencies as 44 kHz, so you'll be able to make decent recordings with any current PDA.

What did I test?

The subjective sound quality with both music and (because Resco has speech-specific, advanced encoders - that is, vocoders) speech. Furthermore, I've also tested how much CPU cycles the apps take under the same conditions. The more CPU cycles required, the less battery life. I also tested Resco's AGC functionality.

Speech

Because VITO and NoteM doesn't have any vocoders, I've compared the quality of Resco's speech vocoders to the standard MP3 recording quality of the two other apps at comparable bit speeds. As it has turned out, in the non-quality area, Resco's Speex vocoder does indeed deliver very good results. It will be worse than the 44 kHz MP3's produced by NoteM, but will still have pretty good quality.

(Incidentally, you may want to check out http://www.speex.org on Speex.)

I've chosen a small part (around 35 secs) of the Lord of the Rings, Book 4 (ISBN: 951-0-27880-7; WSOY, 2003), in Finnish, for the test material. Unfortunately, I couldn't come up with anything English - I mostly listen to Finnish and German audio books.

Music

You may also want to have a recorder that is not only a vocoder. If you know anything about digital signal processing, or just encoding speech, you know that vocoders are pretty lousy at encoding non-human sound. This is why, while a GSM phone (with a GSM vocoder) can encode human speech quite well, it'll miserably fail at transferring music or anything non-speech - the result will be complete nonsense. Therefore, I've also tested and compared the "traditional" codecs of the three programs to see which is the most suitable for this task. As it has turned out, Resco was much worse at this than the two other applications.

I've chosen the tenth song from the album 'Lo! Born Is Our Emmanuel', Laestadian Lutheran Church, Finland, Espoo, 2001. Despite sung by Finns, it's in English. I've turned up trembles in the desktop when playing the song so that anomalies with the trembles can easier be spotted.

CPU utilization

Encoding music/speech requires CPU cycles. The more complex the encoding algorithm is, the more CPU cycles are required - that is, the CPU utilization increases. It means (sometimes drastically) reduced battery life. This is of extreme importance when using a PDA without a charger to record something very long -for example, a several-hour-long session. If you know beforehand that you'll need every bit of the battery, you can also base your choice of application/encoder based on detailed CPU utilization data.

All the results are measured on a 'clean', freshly hard reset iPAQ 2210 (400 MHz PXA255), with XCPUScalar 2.82.

Automatic Gain Control (AGC)

Resco supports automatic gain control. This may be known for anyone that has ever seen a consumer (read: cheap), portable, battery-operated tape recorder that (mostly exclusively, at least in the cheapest price category) has AGC to set the recording level automatically to both avoid distortion for excessing the 0dB threshold and, also, to amplify quiet sounds to be well above the noise level, which is pretty high with Compact Cassette-based systems.

In 16-bit recording (which means 96 dB Signal to Noise Ratio (SNR) and dynamic range, as opposed to the 45...55 dB SNR and 52...62 dB dynamic range of not-the-lowest-end Compact Cassette tape recorders), the need for AGC is much less than with both, SNR-wise, low-quality (Compact Cassette-based) analogue recorders and/or 8-bit A/D encoders. This means you don't need to compress the dynamic range of a 16 bit source because it won't make the more quiet speakers sound more intelligible.

Unfortunately, the sound units of the current PDA's aren't able to deliver 96 dB SNR. I'd say, without actually measuring it (just by depending on my ex-audiophile ears) that they're around 40-45 dB - that is, they can even be more noisy than a simple tape recorder (I mean a tape recorder that uses high-frequency and not constant-current/magnetic deletion. The latter only deliver 20-35 dB SNR). This also means it's pretty pointless to use any kind of AGC on a current PDA: the sound input hardware is inherently so noisy that the 96 dB SNR of the 16-bit encoding is way, orders of magnitude higher than the SNR of the sound input units - the mike, its amplifier unit etc.

Please note that much as VITO also has a gain control slider, it is NOT AGC - it just increases/decreases the pre-amplification, just as with some PDA's System/Sound applets. Also note that turning this up will result in problems. Unlike with analogue technology, you can't go over the 0 dB threshold when recording/digitizing stuff without distortion. With analogue technology, you can (however, the additional, post-0 dB dynamic range is only some 3....12 dB's, depending on the tape type and quality). This is why you shouldn't turn up your gain control to the max with VITO.

So, is it worth using the AGC in the Resco at all? To find this out, I've made three recordings of a quiet sound source (the same Finnish LOTR as with the sound test). I used SPX, in the recommended 32 kHz/q3 (18 kbps) mode. As opposed to the regular speech tests, I waited some 5-10 seconds before starting to play the stuff so that the AGC can have time to 'kick in' and to reach its maximal amplifying.

The according files can be found here. As can clearly be noticed, the recordings with middle-level (50) and max-level (100) AGC level deliver no additional legibility, just very annoying noise level increasing during the pauses.

That is, you won't gain anything if you use AGC (or, for that matter, any kind of manual gain control) - quiet speech won't be more intelligible / won't stand out more from the noise introduced by the PDA hardware. It's just that you won't need to turn up the volume level on your amplifier that much.

Inherent noise of the encoders

The next test was to find out how much inherent noise Resco's SPX codec has, as opposed to the MP3 codec of NoteM/VITO, its most important alternatives. As has turned out, the SPX codec codes quiet speakers as good as NoteM. This means SPX has no more inherent noise than NoteM's MP3 encoder. This is certainly good news.

You can also download the NoteM and the Resco recording so that you can hear the difference yourself.

Scheduling

Both VITO and Resco offer scheduling capabilities. However, neither of them work OK without the well-known "wake up PPC to full power mode" Registry hack (see http://www.pocketpcfreewares.com/en/index.php?soft=876 or http://tweaks2k2.com/downloads.htm). Unfortunately, the help files / documentation of the two apps don't mention this - this is surely a problem with the documentation of both apps because non-expert PPC users won't have a clue about it. Without the hack, on WM2003 devices (WM2003SE included!), only 15 seconds are recorded.

Screen off

Both VITO and Resco offers the screenoff capability during recording (and playing). NoteM, unfortunately, doesn't, only during playing.

VITO has a drop-down list in its Settings menu to set when the screensaver should 'kick in'; it has no screenoff icon, unlike Resco. It doesn't switch on the screen if you (accidentally) press the D-pad or the Action button; it only wakes up by pressing one of the application buttons. If they're unmapped, then, it'll dim the screen after the set interval; if they're mapped to start some other programs, however, the screen will stay on.

Resco is certainly much more configurable. It has three modes:

- use any HW button to turn the display on immediately (that is, it works the same as, say, Pocket Windows Media Player's (PWMP's) display-off method)

- the same as above, but the arrow keys don't turn the display on

- advanced protection: it lights up the screen so that the user can tap a specific region. If there's no tap at the designated screen area for 5 seconds, the screen dims. If there is a tap, then, the same procedure is done once more, at the other corner of the screen. This is by far the best protection against accidental screen-ons, which would otherwise result in faster battery depletion.

However, I still think using an external tool like ScreenLock (http://www.freewareppc.com/utilities/screenlockforpocketpc.shtml) is the best solution for dimming the screen and locking all the buttons (including the power button) because it locks all the buttons and, after switching the screen back on, you don't need to tap the screen twice.

Offline file conversion

Both VITO and NoteM supports offline WAV -> MP3 conversion. Unfortunately, Resco doesn't. It'd be great, however, especially because of its OGG encoder.

Although both NoteM and VITO lists two bit speeds, both will default to only one - 128 kbps with NoteM and 64k (with 22 kHz sampling frequency) with VITO, no matter what you choose (56 kbps or 96 kbps/44 kHz with NoteM and VITO, respectively).

As you may have guessed, VITO's offline encoding sound quality is very bad, while NoteM's quality is astonishingly good. The original WAV file's MP3 version (the same file used with the music test), encoded with MP3Producer 2.35, VBR, 128...256 kbps, 44 kHz, Stereo, can be found here.

Memory cards

It's also worth examining how much screen taps it takes to reconfigure the output directory. I've been having problems (7 screen taps to reconfigure the app is way too much) with NoteM's behavior because of its inability of dynamically changing back to the built-in RAM memory when the memory card is taken out. This subject may be important for you if you want to record while your card is taken out for, for example, a backup.

Resco automatically switches back and forth between using the main memory if you take out/put back the memory card. (Configuring the output is 6 taps, but, because of the just-described behavior, you'll rarely need to reconfigure anything.)

VITO also does this automatically in the CF card -> main memory direction; in the opposite direction, you have to choose the memory card again. It's 2 screen taps only, so not a big hassle.

NoteM is a clear loser in this: no automatic reconfiguration takes place and the manual reconfiguration takes 7 taps.

Long-time recordings on memory cards

With NoteM, it's important that you format your flash ROM cards from time to time with either a PDA app capable of doing this (Resco File Explorer, Pocket Mechanic, Storage Tools etc), or, in a desktop PC (format drivename: /q /u) because, with a card that has been heavily used and not formatted after that, you may end up with long recordings that have serious skipping problems, especially after the first hour.

Stability

I've also scrutinized the overall stability of both NoteM and Resco (the two most usable recording solutions). Unfortunately, there may be problems with both of them during casual use; mostly with NoteM.

Freezing, non-closed files

Unfortunately, if you press the Stop button while recording, NoteM sometimes (in about 1-3% of the cases; then, it just further displays 'Recording' and the CPU utilization jumps to above 90%) just refuses to close the file on the storage card, leaving a simple 138-byte-header on the card. The file can be fully restored with using chkdsk though, explained in the next section. It's a good idea not to record anything else on the card though before the checkdisk operation, though, if you don't want to get messed-up contents.

You really should avoid doing the following during recording to minimize cases like this:

- do not run other, CPU/memory-intensive tasks during recording. If you really have to, restart the recording before starting them and after finishing them, so that the loss can be minimized if NoteM refuses to close the files.

- no not insert/remove memory cards to/from the other slot

Resco also has problems sometimes, especially with non-freshly-reformatted cards, if you stop a recording and start another one too quick, without waiting at all. Then, the current file will remain 0-byte-long and can't even be restored with chkdsk (meaning it's lost forever). If you wait at least 3-4 seconds before starting the new recording, this will not really happen. And, always remember to format (format drivename: /u /q) your cards regularly if you use both Resco and NoteM. With a freshly-formatted card, I haven't managed to reproduce the problem, unlike with a card formatted weeks before.

Memory cards/computer shutting down on critically low battery level

Some PDA's (for example the iPAQ 2210) automatically detach memory cards when the battry charge level decreases under a given threshold (with the 2210, it's some 9-10%). If you make long recordings and detaching the memory cards happens during that, it's still good to know whether you can safely retrieve the recorded contents from the card after a check disk (the chkdsk /f drivename: command) on a desktop with the card inserted in a card reader. This restores files lost during powerdowns like this. (Please note that, although there're PPC-based chkdsk solutions - Pocket Mechanic, Storage Tools - for the PPC, neither of them worked as well as the desktop variants. Sometimes they have even completely messed up my cards, which the desktop chkdsk has never did.) In these tests, I've used an iPAQ at 10-11% charge level to test whether I can retrieve all the contents recorded before the power down, the file is a valid sound file (that is, no post-written header is needed for it to become valid) and even quick repositioning can be used with them.

The other test checked the same with recorded files written straight to the main memory. The PPC operating system shuts down the entire system around 1% battery level. Then, no chkdsk is needed - the already- written file contents will be there. Both Resco and NoteM excelled at this: the already-recorded MP3 and SPX contents were all readable and valid.

NoteM: memory card contents can be restored by chkdsk without problems; affected files are restored on the spot.

Resco: memory card contents can also be restored by chkdsk; they're collected as CHK files, so you have to rename it (them) manually. Fortunately, they have the right timestamp, so it's hard not to find where they belong to.

Playing

As MP3/OGG players, Resco and VITO have some advantages over the built-in Windows Media Player (in addition to be able to play OGG files) - for example, Resco supports in-PDA playlist sorting, which is painfully missing from even the latest PWMP9 builds.

However, looking at the battery consumption figures, I wouldn't recommend any of them as a regular MP3 player. Resco consumes slightly more CPU cycles than the built-in PWMP and definitely more than the absolute killer BetaPlayer 0.50. VITO, on the other hand, has a very CPU-intensive MP3 and OGG decoder - do not use it for playing MP3's/OGG's if you want great battery life!

Other remarks

NoteM 1.21

This recorder is unbeatable for basic quality recording purposes because it's

1, free

2, doesn't take much memory

3, its power consumption is far less than that of the two other apps in recording

4, its 56 kbps/44 kHz MP3 mode isn't worse as the 64 kbps/44 kHz MP3 mode of VITO and is much better than any OGG/MP3 mode of Resco.

However, it doesn't support VAS and has some other bugs like setting the system volume to zero if you (re)play any sound file in it. Re-setting/enabling the system volume again can be quite tiring after some time.

VITO 2005

The other choice for recording quality MP3's. It has VAS (which, unfortunately, consumes a LOT of power! With VAS, forget about using it to record several hours without putting your PDA on charge!) and some other goodies. Its VAS is better than that of Resco because, in a separate, human-readable file, it also stores the exact positions of the offline points so you'll always know when the recording stopped. Furthermore, these index files are human-readable so you won't necessarily need VITO to know where the recording paused.

As with NoteM, it doesn't have any special vocoders (except for the almost-useless system-level GSM) so you'll need to stick to its 44 kHz/64 kbps MP3 mode to record even speech. (As with NoteM, forget the 11 kHz/16 kbps mode.)

A side-remark: the CPU usage of the scroller routine of the VITO Today plug-in is 60% (only runs while you're on the Today screen), so you may want to completely disable its Today plug-in.

You may also want to check out Geekzone's review of the app at http://www.geekzone.co.nz/content.asp?ContentId=3763.

Resco 3.20

This beast is probably the most controversial recorder app. First, in low(er)-quality speech recording, it offers much better speech quality compared to generic-purpose MP3 recorders recording at the same speed. Even with VAS activated, its power consumption (in the recommended SPX, 32 kHz, q3 (18 kbps) mode) is moderate, compared to VITO; but certainly more than that of NoteM, the least power-hungry recorder app.

However, if you want to use it for high-quality speech/music recorders, it's better to forget it altogether. Its MP3 coder is plain bad. It's like the first MP3 codecs on the PC, back in 1996 - or even worse. There're just no trembles (highs) - and, therefore, the samples made with 44 kHz sampling frequency (even in "high-quality" modes at 320 kbps!) are as bad as the 24 kHz samples. If you don't believe me, just listen to the recorded samples.

The MP3 codec is useless for high-quality recordings... what about the OGG coders, then?

Well, they are pretty bad, too. First, on current, 400/520 MHz PXA255/272 devices, you can't use any sampling frequency over 24 kHz, which alone restricts the sound quality. And, even at 24 kHz, the CPU utilization is around 90-95%, which is pretty bad, means huge battery load and the inability to be operated in a background task. (For comparison, if you run the active NoteM in a background task, you'll see almost no speed decrease in the foreground one.) On a 400 MHz PXA255 device like the iPAQ 2210, not even the 24 kHz mode is usable because of the skips.

The OGG built-in recording test at 32 kHz ended with a positive result on my 520 MHz Pocket Loox 720 using main memory (on the 400 MHz 2210 it still ended with negative result), but there were skippings in the recording in this mode. This means 32 kHz recording modes (let alone the ones with 44 kHz sampling frequencies!) are not usable even on 520 MHz devices. Unfortunately, the same stands for 24 kHz modes on the iPAQ 2210: much as the app says it's OK to use them, there will be inevitable skips, even if you don't touch the PPC, don't start other programs etc.

All in all, this app is excellent for recording low(er) quality speech, but is almost unable to record high-quality music because of its sub-par MP3 coder and very CPU-intensive OGG encoder.

CPU utilization remarks: you can shove off some 8% CPU utilization if you always use the file list view instead of the VAS/AGC view during recording. Then, if you use VAS, make sure that the "Use VAS even if the VAS panel isn't visible" is checked in Tools/Settings/General.

Good to know: the SPX player module in Resco has a nasty integer overflow error that makes all SPX files over around 28 minutes to be played wrong. If you move the position to beyond 28 minutes, the contents of the file right at the beginning will be played. This bug doesn't affect recording, though - you may even record 6-hour-long SPX recordings without problems. If you need a desktop-based player that doesn't have the above bug, you may want to get foobar2000 from http://www.foobar2000.org/. It's fully compatible with Speex files.

The Winners

Low(ish) quality speech (under 20 kbps)

The two MP3-only apps are certainly losers in this category: both NoteM and VITO are only able to record with 11 kHz only at 16 kbps. Their 16 kbps recordings are far worse than that Resco utilizing SPX at 32 kHz sampling frequency, with the quality 3 (resulting in a 18 kbps stream). I certainly recommend this recording mode: it's not as good/generic (it's mainly speech-only) as, say, NoteM in 44 kHz/56 kbps mode, but still delivers good results.

I've also compared the quality of the above-detailed mode to the same SPX codec used with 16 kHz sampling frequency. The latter should be avoided; even at q5 (42 kbps), it produces worse results than the 32 kHz/q3 mode.

Interestingly, I've found the subjective quality of the q3/32 kHz (18 kbps) SPX mode better than the q4/32 kHz (25 kbps). The latter was a bit more distorted. Also, the q4 / q5 modes require considerably more CPU cycles than the q3 mode, which "only" consumes around 55%.

High quality speech/music

NoteM (MP3 44 kHz) 56 kbps. It's not worse than the 128 kbps mode (which should only be used when coding existing stereo WAV's into MP3's and not at recording from the built-in mike) and the VITO 64/96 kbps mode. The MP3 modes of Resco are very bad and its OGG encoder, while having good quality, requires just far too much processor power still not available on present-day PDA's.

Music files produced by the apps

Speech files produced by the apps

Legend - the file names

First letter:

r: produced by Resco

v: produced by VITO

n: produced by NoteM

The number+k after this:

bit speed in kbps. For example, the first number in r160kq4 means that it was recorded at 160 kbps.

The optional number+kHz after this: sampling frequency. For example, in v16k11kHz, it means 11 kHz.

The optional q+number after this: quality in Resco. For example, in r56k24kHzq3, it means quality 3.

Generic CPU utilization on the iPAQ 2210 (reading from / writing to SD)

I've listed both the (weighed) lowest and highest CPU utilization I've encountered. I've made sure that the weighing should be done so that the average is just at the median of the two results.

Recording

Vito, 44kHz/96k, VAS off	46-49%
Vito, 44kHz/64k, VAS off	46-49%
Vito, 44kHz/64k, VAS on, no recording	15-28%; depending on the noise level: in a quiet room, lower; with noises just below the VA level, higher
Vito, 44kHz/64k, VAS on, recording	around 95-98% (!!!)
RESCO, MP3, 44 kHz, q3 - 96 kbps	62-68%
RESCO, MP3, 32 kHz, q3 - 96 kbps	50-55%
RESCO, MP3, 24 kHz, q3 - 56 kbps	40-48%
RESCO, SPX, 32 kHz, q5 - 44 kbps	85-86%
RESCO, SPX, 32 kHz, q3 - 18 kbps	55-58%
RESCO, SPX, 16 kHz, q4 - 23 kbps	55-59%
RESCO, SPX, 16 kHz, q5 - 42 kbps	69-70%
RESCO, SPX, 32 kHz, q3 - 18 kbps, VAS:	not recording: 11-13%; recording: 54-58% in File List view; in VAS Panel view, +8%
RESCO, SPX, 16 kHz, q3 - 16 kbps	39-43%
RESCO, RAF, 11 kHz - 44 kbps	9-17%
RESCO, RAF, 22 kHz - 88 kbps	12-17%
RESCO, OGG, 24 kHz - 50 kbps (predefined)	88-94%
RESCO, OGG, 24 kHz, q2 - 35 kbps	88-94%
NoteM, 56k	15-22%
NoteM, 128k	15-26%

Playing

Italic: worst, bold: best

64 kbps mono 44 kHz MP3 (produced by VITO)

Resco	7-16%
VITO	37-44%
NoteM	6-15%
WMP 9	6-13%
BetaPlayer 0.50	6-8%

192 kbps stereo 44 kHz MP3 (produced by Lame 3.92)

Resco	8-32%
VITO	44-49%
NoteM	19-27%
WMP 9	15-25%
BetaPlayer 0.50	12-16%

OGG, mono, 50 kbps, 24 kHz, q3, created by Resco

VITO	34-37%
Resco	12-18%
BetaPlayer 0.50	6-10%

Please note that BetaPlayer may have worse OGG sound quality that the competition, for example, PocketPlayer - check out http://www.firstloox.org//forums/showthread.php?t=3187 for more info.

WMA, 82 kbps, stereo, 44 kHz, q3, created by desktop WMP9 (I've only made this test to find out whether it's worth using BetaPlayer instead of WMP to play WMA files)

WMP9	11-18%
BetaPlayer 0.50	6-13%

As far as the playing CPU utilization is concerned:

- do NOT use VITO for playing stuff!

- prefer using BetaPlayer, it has the least battery consumption with both MP3, OGG (and, incidentally, WMA) files.

Conclusion

There's no clear winner. All the three apps have their strengths and their share of problems. You have to consider the strengths/weaknesses them, taking into account your needs (do you only need to record low-quality speech? Or, on the other hand, high(er)-quality music?). If you want to make (for the used bandwidth) astonishingly good voice recordings, get the Resco. Otherwise, get NoteM or, if you need the special tricks (quick play) of it, VITO. The latter application (VITO) I wouldn't recommend for generic sound recording/playing, especially not if operated on battery, though.