Quality of HDV MPEG Layer II audio

Edin1 wrote on 9/24/2005, 11:13 PM
Recently, I did a comparison of MPEG Layer I, II and III codecs, just to see how much of a difference is there between them, if Layer II sounds better than Layer III (at higher bitrates), and how much progress has been made in between them.
The progress is audible, and MP3 wins.
I did this also to find out how good is the Layer II at 384kbps, which is being used for HDV audio. I have to say that I am dissapointed. Overall, it leaves a different taste, but what is noticeable is that even at 384kbps, Layer II simply doesn't cover the high frequencies almost at all. I am talking about the range between 12kHz and 20kHz. I had some high-frequency hiss and other sounds and noise, and Layer II loses them.
MP3 covers those even at half the bitrate (192kbps) of Layer II.
Quality wise, MP3 would have been a better choice, but I think that Layer II was used because it's easier to edit and synchronize with video.
Personally, I am strongly convinced that plain uncompressed PCM audio should have been left in place, as 1152kbps difference hardly means anything to video, and a world of difference to audio.

Note: I used Sound Forge 8.0 to compress the audio, and Winamp, Windows Media Player 9, and Media Player Classic for playing back the compressed files (whichever player was able to play them).

Comments

farss wrote on 9/25/2005, 12:09 AM
With HDV the focus is clearly on video and I don't think the designers need to apologise for that, it's no big drama to use an audio field recoder, in fact even with most DV cameras I'd suggest doing the same if you're serious about capturing location audio. Another plus is it lets the cameraman focus on what he's doing, life was hard enough for the guy shooting 4:3 SD, 16:9 HD is a bigger challenge.
We did a test, recodering live opera to both digibetacam and HDV and the difference in the audio is pretty remarkable, the audio off the HDV even to my musically challenged ears sounded very muddied, the DB audio sounds good enough to sell. Audio feed was from the in house Euphonics desk so we bypassed the mic pres and mics, about as clean a signal went into both devices as one could get.
Bob.
Edin1 wrote on 9/25/2005, 7:17 AM
That's another point I forgot, and had no time to mention.
Even with uncompressed PCM, audio quality of almost all except "professional" DV camcorders is not serious, because of the inferior audio circuitry. There is low dynamic range and high noise floor, even in my VX2000, and, I heard, in VX2100. External, off-board mics usually improve this (a $30 Audio-Technica condenser mic does a nice job, and cheap), and good preamps connected to line-in improve the signal to noise ratio even further, but a good, separate audio recorder is even better.
Almost every top professional video and film recording uses separate equipment for audio. I don't remember seeing those big film cameras with onboard mics.
Spot|DSE wrote on 9/25/2005, 7:56 AM
Edin
Can you post how you tested? At equal bitrates, using the Z1 and quality mics, (AT 4053) I recorded tone at a distance of 5', and recorded to ATRAC, HDV, PCM, and MP3 on a Rio.

Using spectrum analysis and frequency charts compared to the original, it was immediately clear that the HDV audio flow is fine for dialog, and most audio applications, but I'd certainly not use it for music (but wouldn't use a DV or HDV camcorder for this any way as the DACs usually aren't all that) and found it was a lot better than I'd expected it to be. Of course, the ATRACs recording fell apart at 10.4k, whilst MP3 at 384 slowly fell down starting at 15k, and HDV held solid to 15k, and rapidly fell apart after that. PCM was clean all across, as much as the DACs in the camcorder would permit. Interestingly enough, the DACs in Thomson LDK cam weren't as good as the Sony HDV, and it's 5 times the cost. Sony HDV isn't as good as the Edirol R4, and it's a quarter the cost.
farss wrote on 9/25/2005, 8:07 AM
Is just tone a valid test with these kinds of compression schemes?
I don't know what form of modelling mpeg-1 layer 2 uses but certainly with how layer 3 works tone at around 0dBFS probably encodes very cleanly, it's compex audio with wide dynamics that stress the system.
I think also layer 2 holds up sort of OK if you keep the level up but try adding 12dB gain in post and you could be in serious trouble.
I think we're all saying the same thing, video cameras are pretty bad as audio recorders and the HDV ones are even worse. If you're serious about your audio, buy a good field recorder. Other thing with a good field recorder like the R-4 is being able to multitrack so you can make your mixing decisions in post.
Bob.
Spot|DSE wrote on 9/25/2005, 8:23 AM
The tone test isn't ideal, but it's the only way I could see exactly what was going on, using a sweep. I also used "Money for Nothing" from the Dire Straits, and it showed approximately the same result. Dynamic range of HDV was better than MP3 in lower freq's, and then it fell down starting around 10.5, but not significantly so until we hit 15k. I posted my results on our site.
Edin1 wrote on 9/25/2005, 10:32 AM
That's what I meant. For frequencies below 10-15kHz, Layer II sounds "different", and I wasn't sure if it were better or worse than MP3. But high frequencies were either out, or distorted, even at the highest bitrate of 384kbps.
I compared the 384kbps Layer II with 192-320kbps MP3, and found that MP3 covers high frequencies better, even at lower bitrates.
Below 192kbps, MP3 doesn't sound significantly better than Layer II at 384kbps, but such a difference in bitrate makes using Layer II pointless.
The point I originally wanted to make is that Layer II, simply put, sucks!
MP3 at 256kbps would have been a much better choice; you would get better audio quality (Transparent), and still have 128kbps more for video or anything else!
Spot|DSE wrote on 9/25/2005, 10:42 AM
First, I don't think Sony wanted to license MP3; second, I disagree that MPEG layer II "sucks." I was very, very opposed to it until I started testing it. Paper charts show one thing, and what I'm hearing is somewhat close to what benchmarks show. But for what I've heard, it is adequate. It's certainly not optimal, no doubt. The upgraded bitrate DOES sound better to my ear, it's warmer and smoother, even though you can't "see" that in charts. But it does drop off rapidly at 15.5k, almost like a stone. I don't hear that warmth in the MP3 at lower or even equal bitrates.

Are you testing by merely encoding to the format? Or are you actually using the camera? Because that's a fairly critical component. The compression tools used are important in the process of judging. I tried taking Sound Forge and a PCM file and rendering it to various formats/bitrates. It didn't give me a comparative picture to what the camera itself is doing. Remember, the Z1 can record PCM or MPEG audio, depending on the format you're using. Output that same mic to a splitter feeding an R4 and Rio, and you've got a great 3 way baseline each time you record.
Edin1 wrote on 9/25/2005, 12:03 PM
Well, that's why I've mentioned that my encodings were done with Sound Forge 8.0. Other encoders may do a better or worse job.
But the main point remains; camcorder's onboard sound should be mainly used as a reference (to synchronize externally recorded audio with the one on tape), and alone only when audio quality is not critical, you want to record the ambient sounds, and have nothing better available.
Sony has made the point that onboard sound is not meant for professional recordings by including a lossy and aged audio format to carry it.
Spot|DSE wrote on 9/25/2005, 12:12 PM
Again, I have to dispute that. Audio that is dialog, is certainly fine. Ambiences actually suffer more than dialog because it has a larger frequency range and is lower in signal by comparison. Dialog recorded on the Z1 is being broadcast every day on numerous projects. If you want better audio, then use the LANC to connect to a device like the R4, which is why I bought one, but dialog on the HDV cams (and it's not just Sony, it's the spec) is all in MPEG 1/layerII format.
Have you actually recorded dialog on the Z1, or A1? That is where the real test is, otherwise it's book knowledge vs street knowledge. As I mentioned, merely encoding in Forge doesn't tell the tale. Try it. You'll quickly be able to compare the PCM from the cam and the MPEG from the cam, and view them and more importantly, hear them.
"The Hunting Show" on CMT is using on-board audio from the Z1 every week, btw, they don't use any field mix at all. Just a shotgun and wireless fed by ECM 55
Serena wrote on 9/26/2005, 12:47 AM
Spot, I have to ask: what is an R4? The Rio I know (I believe I know!).

Serena
farss wrote on 9/26/2005, 1:15 AM
Edirol R-4. I bought one at an excellent discount on the last day of NAB, very good unit for location sound and just as a general audio recorder. 4 channels at upto 24/96K, each bank of 2 channels can be line or mic, phantom power or not and switchable limiters etc. The mic pres are probably not studio quality but they sure leave anything I've found in a camera for dead.
Bob.
Serena wrote on 9/26/2005, 1:37 AM
Thanks Bob. Turns out that I'd looked over one in John Barry's. Nice, but not cheap either.

Serena
farss wrote on 9/26/2005, 2:23 AM
If you're feeling adventurous, sweetwater.com in the USA have them MUCH cheaper than the local ripp off merchants. I can vouch for Sweetwater delivering the goods.
Bob.
Serena wrote on 9/26/2005, 2:44 AM
Had a quick peer -- looks about $400 better than Videoguys. Thinking.....

Serena