An AskMetaFilter question says “My grandmother’s first language [Ladino] is nearly extinct. I’d like to record an interview with her for archival purposes; how should I go about it? … I’m linguistically literate, but far from an expert, so advice from anyone with linguistics experience (particularly field lingustics) is especially appreciated.” If you’re a MetaFilter member, you can respond in the thread; if you’re not but have useful advice, post it here and I’ll pass it on. One response there seems important enough I’ll repost it here, in case anyone is thinking of doing something similar themselves: “Use No Compression. Can’t stress this highly enough. Your recordings must be uncompressed. If you record to MP3 or whatever perceptual encoding scheme, you will lose phonetic information.”


  1. I second the “no compression” recommendation. PCM Wave (PC) or AIFF (Mac), 44.1 kHz, 16-bit sample is about right for speech. On Macs, one should be careful to distinguish between uncompressed AIFF and compressed AIFF-C. As for hardware, that’s a different story. Basically any decent recorder that records uncompressed audio will do, but a separate microphone is a must. Some knowledgeable people swear by Audio-Technica’s products, especially their less-expensive lapel mics.

  2. marie-lucie says

    I agree that you do need a separate microphone from the one that is already in the recorder, however good that one is. With a lapel mike, don’t you need one for each person? For instance, if you are the interviewer, don’t you need a mike for yourself as well as one for the interviewee? Or can you trust the inside mike to record you (but it will record surrounding sounds too), while the lapel mike records the other person?

  3. m-l,
    you’re right, a lapel mic is best for recording more or less uninterrupted testimonies with minimal input from the other person (which is the only way I’ve ever used such mics). There are plenty of microphones with stands and windscreens out there (e.g.) that would be more suitable for a true interview.

  4. I’ve been recording some family moments with a small hundred-dollar camera, a Canon A560 Powershot. There are now better cameras on the market with more stability features, etc., but if anyone wants to see what it can do, this is my most popular YouTube video so far, my local town hall health care political meeting:
    The video function works very well even in low light, even in places where you can’t take a photograph without a flash, but the sound quality isn’t all that good for music. The other problem with family videos is privacy. I would love to put them on YouTube to share long distance, but that means either they are publicly accessible (not acceptable for videos with minors or it might raise some employment-type issues like the ones associated with facebook) or you have to go through too many convolutions for family members who are not that tech-savvy. Basically that means burning a disk and mailing it, which means looking for editing software (I tried a YouTube downloader once that had horrible quality), and spending time learning new software skills.

  5. FLAC FTW!
    (It is just the implication that compression is necessarily lossy that I wish to refute; if They say lossy compression is bad I take not the slightest exception.)

  6. SnowLeopard says

    I can’t comment on microphones or linguistic field methods, but Audacity is a pretty good (and free) audio recording and editing program. Documentation seems to be sparse or non-existent, but I was able to figure it out my own with a little experimentation.
    I use it to convert language courses from cassette tapes into something I can listen to on my ipod (which requires a separate downloadable plug-in because mp3 format is apparently proprietary), but the default sampling frequency is CD quality.
    In terms of content, the linguistic data is obviously vital, but I would also try to extract as much historical and biographical detail as possible. When I was interviewing my father about his childhood and teenage years as a refugee and more in the years during and following World War II, he mentioned startling details that I’ve been unable to find documented anywhere else.

  7. Peter Austin says

    Never record directly to a computer as they generally do not have good quality sound cards. Use a digital recorder that records in uncompressed WAV format, never mp3 and buy one or two decent microphones (see Monitor the recording with headphones on to make sure the volume is set right and the machine is recording properly. Pay attention to the environment and avoid noise, eg. fans, refrigerators, turn off the TV, keep away from windows and reflective surfaces, avoid traffic noise (that’s also why you need to monitor the recording).
    If you want ideas about how to do fieldwork get a copy of Claire Bowern’s book (published by Palgrave Macmillan and available through Amazon). Make sure you prepare before the interview so as to maximise the time available with your speaker.

  8. Well, hmm. This is kinda interesting… Is the problem that conventional psychoacoustic modeling is a blunt instrument that sets the “can’t hear the difference” threshhold too low? Or is it just wrong?

  9. I mean… sets the bar too low, sets an ‘audible difference’ too large… You all know what I mean.

  10. Nijma – I had thought that you could have a YouTube channel that was only available to invités?
    Has this changed?

  11. a YouTube channel that was only available to invités
    Yes, but I believe it’s password protected, requiring a certain level of technical savvy to set it up and use it. Can someone elderly who has just learned how to email do this? Then there is still the problem that it is out of the control of the user. There have been all kinds of instances of information shared by an entity, accidentally or not, when it was assumed by users to have been private. Who can you trust? Also YouTube can take anything down when they want, as they often do with copyright claims; other services have gone out of business completely, sometimes notifying their users and sometimes surprising them when everything stored “in the cloud” is lost.

  12. I think I have posted this link before, but Mor Karbasi sometimes sings in Ladino. |MySpace| |Website|

  13. It is just the implication that compression is necessarily lossy that I wish to refute
    Good point, des. If compression were necessarily lossy, ZIP formats could not exist.
    It is not even necessarily the case that something is lost when analog data are compressed. A sine wave can be compressed into three numbers: amplitude, wavelength and initial value. A sum of n sine waves is compressible into n*3 numbers.
    And of course WAV itself is lossy – not because of compression, but because it involves sampling.

  14. Sampling is itself a compression technique, though not one that most people think of as compression.
    Sampling is an attempt to record data in such a way that,

    1) when it is reproduced, the difference between reproduction and original is below the discrimination threshhold of some monitoring device (such as human being looking or listening)
    2) the recorded data take up less space than the original, as defined for some measure of “space”

    That is also a description of compression.

  15. marie-lucie says

    I am not familiar with the type of equipment needed for acoustic phonetic analysis (eg looking at the formants of individual sounds), but it is possible that the wrong microphone would result in loss of some of the features. I would take Peter Austin’s advice seriously: it is worth investing in a good recorder and microphone, the latter made for clearly recording face to face conversation and interviews rather than for amplifying a speaker’s voice in a large hall.

  16. Uh, oh… technical issue looms in the middle distance…
    In fact, sampling is not the problem. It’s quantization. Those three numbers that define a sinusoidal signal are real numbers, with infinite precision. It’s true that in the modern world of today, one assumes that all signals are quantized, but back in days of yore, there was a non-ignorable difference between bandwidth and bit-rate.

  17. Never record directly to a computer as they generally do not have good quality sound cards. Use a digital recorder that records in uncompressed WAV format.
    Let’s not get carried away by all this fashionable digitalization crap. To record speech for acoustic phonetic analysis later, you have a simple choice:

    1) Use a good, moderately priced tape recorder
    2) Use an expensive digital recorder with a high sampling rate

    In both cases you need a good, expensive microphone, otherwise none of the other hardware is worth the investment.
    Every transition between analog and digital involves loss, when speech is involved. The only question is whether the loss can be discriminated by a given monitoring device (person). You should test this yourself before buying any hardware.

  18. In fact, sampling is not the problem. It’s quantization.
    Sampling is quantization, quite apart from “infinite precision”. But I get your point.
    “Infinite precision” is a tricky notion. It is more like a statement of intent, not a goal that can be reached. Physicists/engineers apparently often think of “precision” as something objective which they try to approach ever more closely. But, for a pure mathematician (outside the fields of numerical approximation), there is no such thing as “precision”. Numbers are numbers.

  19. MattF: I initially missed the fact that you had linked to the WiPe on Nyquist-Shannon. “Shannon’s formulation” there contains the point I was making about discrimination threshhold:

    If a function x(t) contains no frequencies higher than B hertz, it is completely determined by giving its ordinates at a series of points spaced 1/(2B) seconds apart.

    The frequency range which can be discriminated by the “average person” defines how much you have to pay for sampling hardware/software.
    As Peter Austin said:

    Never record directly to a computer as they generally do not have good quality sound cards.

    That’s because a good quality sound card has a high sampling rate. All digital formats including WAV build on that. WAV itself doesn’t give you good quality.
    But you don’t need to pay for a high sampling rate if you use a tape recorder. I claim that is the device of choice for phonetic analysts.

  20. My only point was that once you are sampling faster than the Nyquist rate, the reconstructed function is exact, not approximate. If you’ve lost information in the sampling process, it’s because you’ve quantized the sampled data.

  21. If you’ve lost information in the sampling process, it’s because you’ve quantized the sampled data.
    Are you using “quantized” in a special sense here? All I can see is that if you’ve lost information in the sampling process, then you must not have sampled often enough. I’m not familiar with “quantized” as a synonym for “information loss”.

  22. Surely you don’t mean something like Planck quantization ? That would really be a “technical issue looming in the middle distance”. It might even mean you believe that the world is “continuous”. But that is an artefact of classical-physics thinking, on my view. I’m not a physicist, but it seems clear that we would belong to different camps – and all that has been argued at great length elsewhere, elsewhen.

  23. It’s true that if you sample at less than the Nyquist rate, then you’re not going to have enough data for an exact reconstruction of the signal you’ve sampled– so in that case, the sampling rate would be the problem.
    But audio bandwidths are small– we’re talking 10 to 20 kilohertz– so, the sampling rate required to get enough data to capture all the information in an audio signal is also small. The problem is that, when you do things digitally, the true value of each sample is approximated by a finite number of bits (i.e., it’s a quantized, digital quantity rather than an analog, continuous quantity). So, any signal reconstructed from digital samples is also approximate, and consequently, you’ve lost information.

  24. And I’ll note also (and probably confuse the issue by noting) that analog sampling is no panacea– just a different set of problems, e.g., distortion and averaging.

  25. Might not signals beyond 20kHz, inaudible though they may be to (most) human ears, be relevant for phonetic studies?

  26. Ah, now I get you. Digitalization is inherently lossy, because one has only a finite number of bits at one’s disposal. Not because of measurement problems, but of representational problems.
    In general, as you say, sampling at a higher frequency is not going to get you off that hook. An increase in knowledge doesn’t change the fact that there are only so many trees and hard drives available for recording purposes. Information that is not distributable might as well not exist. I’m reminded of Wittgenstein’s (implicit) argument that a “private language”, being by definition unintelligible, might as well not exist.
    All the more reason for people to use tape recorders.

  27. And I’ll note also (and probably confuse the issue by noting) that analog sampling is no panacea– just a different set of problems, e.g., distortion and averaging.
    With an appropriate definition of “change of medium”, every change of medium is entropic. There is loss when signals are “transferred” from air, through a microphone, to a magnetic tape – even though everything is analog, and there is no intentional compression.

  28. This thread is distinctly non-Hattian, if you ask me.

  29. I think Hat enjoys the sport aspect. It’s at times like this that he has commented: “I don’t understand any of this, but I’m happy to sit back with a beer in my hand and watch the ball being knocked back and forth”.
    John, I have been expecting you any minute to intervene with a point about proto-Dravidian or Chinese. Those are things about which I know nothing. Perhaps this is a good opportunity to explain what proto-Dravidian is ? At the very least, it seems to be some kind of running joke.
    And I bet you’re the down-home kind of guy who prefers a tape recorder. All the right people do.

  30. I think Hat enjoys the sport aspect. It’s at times like this that he has commented: “I don’t understand any of this, but I’m happy to sit back with a beer in my hand and watch the ball being knocked back and forth”.

  31. Stu, I thought that everyone knew that the language of the Garden of Eden was Dravidian, if it wasn’t Dutch, and that all other languages are debased versions of this, the original and true language of man and womankind.
    Others say Basque or Yukagir, but they are detestable and antiscientific.

  32. Gosh, I’ve never given much thought to how Eve communicated with the snake. I had imagined that it was touchy-feely, rather than verbal.
    Is it possible that she used sign language ? Or would that have required a semiotic sophistication hard to credit ?

  33. To go back to the original topic (of decagon recording his grandmother), here are a couple of things to keep in mind other than the recording device.
    . If decagon’s grandmother hasn’t spoken the language for a long time, asking for paradigms, translation, and that sort of thing is likely to be counterproductive. It’s hard enough for fluent speakers. Rather, the first priority as part of making a documentation with someone who hasn’t spoken the language for a long time is creating an environment where she’s comfortable talking the language. Decagon’s going to be better at that than an outsider.
    . Does she have any friends who might be speakers? Getting several speakers together to talk can be very helpful in getting the linguistic juices flowing (Sally Thomason and Marianne Mithun have talked about this).
    . Don’t get discouraged! it might be hard for everyone initially. Don’t worry if a lot of the early interview is in English (or whatever language you commonly use together) – you can start things going by asking for the Ladino words for things that come up in the interview, then building larger sentences and going from there.
    . If she grew up speaking it, probably the easiest things to start with would be memories of growing up, describing her memories of the town she grew up in, and so on.
    . OLAC unfortunately has terribly opaque documentation about how to become affiliated with them. Another option might be a university with a substantial Judaic studies program (here’s a list pulled off google:
    . I’d say you wouldn’t need to do a lot of linguistics in advance, but do have a set of things in mind that you’d like to ask about, and asking for clarification for things you don’t understand will also go a long way to documenting the recordings.
    . Speaking of documenting the recordings, it’s a good idea to keep a record of what you talk about, the date, who’s there, and so on. (This is called ‘metadata’.)
    . Regarding documentation of Ladino, I don’t know, but I’d suggest anything your grandmother wants to talk about will be worth recording.
    Good luck! Sounds like a great project
    Feel free to email me.
    (I don’t have a metafilter acct but I do read languagehat)

  34. I will add to Claire’s excellent suggestions:
    – Content: I absolutely agree that trying to elicit verb forms, etc, in an abstract manner is a turnoff for even a very fluent speaker (they want to actually speak, not recite grammatical forms). If in doubt about what to start with, one subject which is always good for starters is food! Everyone has a largish vocabulary connected with the topic, which is also sure to bring back memories of childhood, siblings, grandparents, etc. Asked for a word, the person might not remember, but the word might pop up in a sentence later if the person’s mind if focused on a reminiscence which brings up the word naturally. For instance, the person might not remember how to say “to fry”, but remembering her grandmother preparing some fried food or other might bring back the word without specific prompting.
    – Asking someone to translate sentences can be a turnoff, and it can also lead to unnatural speech as the person will often strive to imitate what you have just said. It is better to describe a situation and say “what would you say in such a case?” or “How would you answer?” “What if you wanted to answer No?” etc.
    – Let the tape run and record both you and your grandmother. Don’t worry about a lot of English, as it will provide context without which you might not find your way on the tape later. Try to listen after each session. If you only record and don’t listen for a long time, there will be a lot that you can’t recognize afterwards. You don’t have to transcribe everything right away (once you get past individual words and short sentences, it can take a lot of time), but the more you listen, the more you will eventually recognize. Also, if there are things you find impossible to recognize or transcribe, you can ask about them at the next session. But you don’t want to be left with a whole bunch of untranscribed tapes and little idea of what is on them. This is especially true if you have more than one person – in a group there are always some interruptions, people starting to talk at the same time or too fast, laughter, etc which make understanding very difficult, especially long after the session.

  35. Food and cooking are also great fun for ESL conversation, certainly in Taiwan anyway. “How to cook your favorite recipe” or “How does your mother cook your favorite recipe?” were big winners.

  36. Thanks very much for those great comments, Claire and m-l! I’ll add them to the MeFi thread.

  37. Some of my best classes have been with food. One student even brought an electric frying pan and made African plantains complete with chili powder. It’s perfect chance to teach imperatives and increase their vocabulary, since food generates an enthusiasm that makes learning more effortless. I do have a family member on video making lefse, but unfortunately not talking about the cooking process, although I got some other interesting conversation. I guess when people see a tiny camera they don’t associate it with sound.

  38. Gosh, I’ve never given much thought to how Eve communicated with the snake.
    Two(2) of the exceptionally reformed Dutch churches split definitively with each other over the question of whether the serpent actually spoke or merely insinuated non-verbally.

  39. Ah, the joys of Calvinism…

  40. My favorite Dutch sect is the Indifferentists. but I can’t find much written about them.
    1/4 of my ancestors were 19thc Afgescheidenists.

  41. Possibly also called Ahithophelianism, for obvious reasons.

  42. My favourite sect was the Seventh Day Absolutionists, founded by a friend who explained that the Good Book says the Lord worked for six days to create Heaven and Earth and on the seventh day he rested.
    “Doesn’t say anywhere he started work again,” my friend explained.

  43. marie-lucie says

    Doesn’t say anywhere he started work again,”

  44. says

    A latecomer here. I have twice borrowed my DAT (that was about 15 years ago, I know we have something better now) to people doing fieldwork in Siberia. They both came back with like 5 minutes of recordings. It was not that they couldn’t record more. It was just “not important”. I just cried.

  45. What a sad story. Why were they doing fieldwork if they didn’t think it was important?

  46. marie-lucie says

    Maybe their fieldwork was in anthropology or even something completely different (geology?), not linguistics, so for their own purposes they did not think it was important to record the language.

Speak Your Mind