Very Strange Speeds...

[elfs]

I just got back from lunch (500 words of hot man-to-man action! Bwahahaha!) and I noticed something a little weird. I put the latest lesson of my Japanese courses on my laptop in MP3 format to listen to them when I have a chance, and this one came out weird. I'm not sure what's different about the encoding, but instead of the usual 44100/128/2, this one reported itself as being 48000/128/2 and it sounded way too fast. So I ran it through a resampler, trying the difference between 44100 and 48000, or about 0.92. It still sounded too fast. Taking a stab in the dark, I wanted to know what it sounded like at 0.8. That sounded just barely too slow. After 20 lessons I know the narrator's voice.

Pattern recognition set in. 0.92, significantly too fast. 0.8, barely too slow. Slowing down the original stream by 0.844, or 0.92 squared... perfect. What the?

I'm sure there's some audio geek out there who can explain to me what I did wrong with the encode, and why the sound came out sounding too fast.

---

Given that I'm more than halfway through the first semester of the audio lesson kit, I decided to take another stab at a written language kit. I have two to choose from, Japanese for Busy People, Kana Edition, which looks more mainstream, and Japanese in Mangaland, which promises to teach much with much more intensity since manga is kanji-heavy. The advantage to the latter is that I have a lot of manga, most of it without translations.

I'm not impressed with Japanese in Mangaland. Yes, the lessons are as intensive as promised. Yes, you do have to do the homework. But I've run into two typos already that make me think this was a rushed product into a crowding market.

The first typo: "A period is written with a small circle and the coma [sic] is written upside down."

The second typo is much more signficant. In a section on onomatopoeia, a woman's laughter is characterized as "ehahaha." But the hirigana is clearly "ahahahaha," and if they're going to use romaji they should be at laest as precise as a standard romaji text. Screwing up a reader's habits of pronounciation is generally a bad thing.

Worse yet, if you visit their website, the PDF file for "Corrections to Lesson 1" is the file for corrections to Lesson 2-- and so is the file "Corrections for Lesson 2"!

Frack. The technique looks worthwhile and the content interesting. But my confidence is shaken by these mistakes and problems.

Comments

[creepingcrud.livejournal.com]

I'm pretty sure the first of those numbers measuring samples per second; if you ripped at a higher sample speed without having the source be different, it'd be faster, as you report.

[elfs.livejournal.com]

I grok that much: it's sample rate/encoding rate/channels. For me, the mystery is why did the square of the ratio between the encoding rates turn out to be the solution to the speed problem?

[cabrius.livejournal.com]

It sounds like two resampling errors have occured, the effect of which would multiply together since they're essentially 'percentage-of' operations.

A stream recorded at 44.1kHz but tagged as being 48kHz would account for one error, at least. I could see it happening if the software requested 48kHz sampling but the card couldn't handle it and silently remained at 44.1kHz, or failed to switch modes, or such.

I'm not sure where the second error would creep in though, as you'd think any further operations would just propagate that initial error and not introduce any new ones. It could be something arcane within the encoder or decoder, but I'm not too familiar with MP3 specifics.

What about this

[(Anonymous)]

How's this sound: The encoder thinks incoming audio is 48khz (actually 44.1). It resamples the input to 44.1 (first error) but then tags the output file as 48khz (second error).

[poetpaladin.livejournal.com]

Maybe they misspelled, and they're going bankrupt so they don't have time to correct it, and it's really Collections for Lesson 2.

Or not.