Portable Audio Recorders and Drift
 

I have a Zoom H2 and, more recently, a Olympus LS-10. When recording with both units, I understand that they won't be lock-sync with the video, but it seems to me that the drift on the LS-10 is a lot worse (about 1 frame every 4 minutes). I'm wondering if other users have this same experience....just so I know I didn't get one with a screwy internal clock.
Thanks.

Are you recording to wav, mp3, or wma format? For the most accurate clock record to wav file format. The compression to mp3 or wma and the subsequent decompression can do strange things to the clock and increase the tendency to drift. Speed changes that are imperceptible to the ear can have profound effects on sync over time.

I assumed as much, but haven't had the chance to test it out. New project for the day.

I might add, the clocks in consumer mini-recorders are not the most accurate in the world to begin with. It's one of the places where manufacturers cut costs. As long as it's consistent, it really doesn't matter if it's precisely on-spec or not when used for the purposes for which such recorders are designed - recording and playback of personal voice notes and music. Even when transferred to the computer for playback such as an MP3 might or over to a CD, a fraction of a percent clock error is going to be unnoticable. 1 frame in 4 minutes is an error of only 0.01% or 1 part in 7,200. You'd never hear that in a song. But when you need to sync to other devices, now you've got a problem. You need to have uniform adherence to standards in both the sound recorder and the camera. If either is off - if what the camera thinks is one second isn't exactly identical to what the sound recorder thinks is one second - they will drift apart. 1 frame in 4 minutes is a bit much, 1 frame in 15 minutes is more to be expected.

One way to do away with the drift is to redigitize the signal. Take analog out from the recorder and record it back to the edit system at 16/48. There should be no drift in analog out, it is using the same ADA converter clock. If the audio card in the edit PC is any good, nobody is not going to notice anything.

Thanks for your replies. I agree with the fact that these are consumer devices and their clock-sync will not be spot on, but I'm still disappointed in their performace relative to cost.
I also own 2 iRiver units that record line-in with direct MP3 recording. I haven't had sync issues with either on programs running as long as 30 minutes using bitrates as low as 96kbps.

Having just tested the Zoom H2 and Olympus LS-10 with a 23 minute WAV file (Dream Theater's "A Change of Seasons" to be precise) here are the results I got.

Zoom H2 MP3 256kbps. Starts to drift at around 3 minutes, but never goes more than 5 frames off through end of song.
Olympus LS-10 Drifts before 2 minutes, continues to do so ending up 27 frames slow at the 23 minute mark.

As for re-digitizing through the analog input, Its a workaround, but not something I'd relish doing with 1 or 2 hours of material. Defeats the purpose of solid-state memory.

Anyone have better results? Should I try to exchange for another LS-10? Other than the drift, I prefer this Olympus unit to the H2 for build quality, ease of use, quick start-up, and battery longevity. (plus built in memory, speakers, and tactile controls).

I'm curious as to the measured drift (or lack of it) you've achieved with the iRivers. I don't doubt your word when you say you've had no sync problems with them but that's somewhat relative. For instance, broadcast standards, as exemplified by the PBS Redbook technical standards, allow for a maximum sync deviation of no more than +/- one-quarter frame from dead-on at any point in the program.

The 5 frames in 23 minutes you got with the Zoom is still far from acceptable. Granted it's going to be a lot easier to correct than 27 frames though.

Have you tried repeating the experiment recording to wave instead of mp3? Many recorders lose accuracy with the clocking when recording mp3 and the conversion to wave on ingest into your editor can add to the problem.

Yeah. You can't have a serious conversation about sync and drift etc and then say you are recording in mp3.

Can you really refer to it as 'drift' when these recorders were not designed to be locked to anything? The internal clocks on both units are fine, you are trying to use them for something they were not designed for.

"Have you tried repeating the experiment recording to wave instead of mp3? Many recorders lose accuracy with the clocking when recording mp3 and the conversion to wave on ingest into your editor can add to the problem."

Not to mention that once you've recorded in MP3 you gain nothing by converting back as MP3 is a lossy format. john.

I have to agree with John - "drift" is not accurate in describing the sync errors that we must correct for when using these inexpensive recorders.

Much of our language for discussing this comes from the dark ages of as much as 20 years ago, when analog recording ruled.

Drift is what you get with an unstable clock. Drift is what you get when you're using mechanical devices to spin reels and capstans so as to draw a tape past tape heads, with some resulting wow and flutter. Drift of analog recorders was corrected by the two-step process of recording of synchronization signals (for us, timecode) from a master clock, then later reproduction on a player with a servo mechanism that can be bumped faster or slower by a timecode synchronizer that is "resolving" timecode to create "synch lock" between two or more players. Complicated, expensive, and painful when it didn't work.

Now, even the cheapest consumer digital audio recorder has a very stable clock indeed. The error we deal with is that most camcorders are *very* compliant to 29.97 video frames and 48,000 audio samples in one second, but consumer/prosumer audio recorders may use clocks that drive the audio sampling rate slightly above or below 48KHz. Then, when we try to "resolve" our samples to playback over time within the NLE, the computer plays back 48,000 samples in a second no matter how many were laid down during recording... so we get issues with our audio being a little longer or shorter than video in a long take.

Properly speaking, when a second isn't a second, we should call this "timebase error".

Correction of timebase error in audio (resampling) has been covered many times in this forum, some NLEs will allow you to apply a correction to an audio clip, other users will have to pull their audio into a proper sound editor such as the freeware open-source Audacity to apply the correction.

Given that this error typically is noticeable only on long takes, such corrections aren't very timeconsuming.

Very true - but you have to convert an mp3 to WAV or some other audio format such as AIFF when you bring it into the timeline to sync it up to the video in your NLE because the NLEs can't work directly with mp3. Also, not all NLEs handle the sample rate conversions to the 48kHz video standard necessary when importing files recorded at 44.1 equally well either.

IMHO The only legit test of the recorder's suitability for use in recording double system sound for video is to record the original at 48kHz, 16 or 24 bit, wav or bwf. Sure a mp3 recorded to a 2 gig SD card gives about 17 hours of recording time while 48kHz wave fills up the same card in 2 or 3 hours depending on bit depth but so what. Buy a couple more cards - they're cheap.

MP3 is the Devil's work - it sounds horrible and should never, ever, be used for recording.

It should only be used for final delivery of audio where quality is not important.

And even if they are spot on, it doesn't help - it's the video camera 'clock' that has to be the master reference in this situation, even if it's the one that's wrong!

I've found the best solution is to re-sample the remotely recorded audio as 48KHz .wav files, if necessary, to match the camera audio format, and then use something like this procedure

My Video Problems :: View topic - Synchronise external and camera audio tracks.

to re-sync the audio. There are of course loads of different ways to achieve the same results --just adjust the procedure to suit your own audio editor.

Having been gone for Thanksgiving, I wasn't able to retest using the suggestions already mentioned.

I do agree that 'drift' is a poor term to use for timebase error (but its much shorter), and that MP3 is a terrible format...for music. However, I have yet to come across a client who can tell when a voice recording was made compressed vs. uncompressed.

As I mentioned, the thought of re-digitizing in real-time leaves much to be desired. I often record 2 or more hours from the audio board and that would be more time spent in the pre-edit stage.

My intent was simply to see if someone else with the Olympus LS-10 has a similar issue. But thank you all for the lively discussion and suggestions.

Well, the least compressed MP3 formats can not be distinqushed from 16/44 by ear. Countless videos and even indie movies have been shot with MD ATRAC audio, and best MP3 is better. Good MP3 is better than any analog system (legendary Nagras) and that used to be the gold standard of audio still few decades ago.

Better avoid generalisations like "it sounds horrible". I would bet you can not tell apart best MP3 (or even second best) and 16/44 or 16/48, John, with all respect.

Besides, some DVD formats have less than 100 kbs audio stream per channel and people make and buy them by the millions... They would not if it souded horrible.