|
Post by johneppstein on Sept 5, 2018 13:10:06 GMT -6
The thing is that precision hand work is always much slower and therefore more expensive. A lot more expensive.
And manufacturing something like heart stents (not "stints", gotta watch the autocorrect in technical stuff), as high precision as it may be, doesn't come close to matching the tuning of top quality mic capsules. And even so, as Christopher noted, the rejection rate is huge. And requires a lot of human intervention. Fortunately, extruded plastic is cheap and probably the material itself doesn't need to be quite as homogeneous as the mylar or pvc in mic diaphragms.
I think when we talk quality we need to specify what particular quality we are trying to quantify for instance; I’ll hazard a guess that Neumann would tell us the machine tightened capsule is farm more uniform in tension, but I’ll bet the old hand tensioning was far better at compensating for imperfections in the diaphragm material. I’ll also bet If the old Neumann guys could take a CNC machine back in time for milling back plates they would! The key here isn’t is automation better than handmade, it’s like oh so many things the right tool for that application. Well, Klaus says the current tolerences are wider than the vintage. Why that is is a good question - it could have to do with the machine construction or it might just be that Sennheiser has loosened the tolerances to reduce the number of rejects and save money.
|
|
ericn
Temp
Balance Engineer
Posts: 15,019
|
Post by ericn on Sept 5, 2018 13:15:32 GMT -6
I think when we talk quality we need to specify what particular quality we are trying to quantify for instance; I’ll hazard a guess that Neumann would tell us the machine tightened capsule is farm more uniform in tension, but I’ll bet the old hand tensioning was far better at compensating for imperfections in the diaphragm material. I’ll also bet If the old Neumann guys could take a CNC machine back in time for milling back plates they would! The key here isn’t is automation better than handmade, it’s like oh so many things the right tool for that application. Well, Klaus says the current tolerences are wider than the vintage. Why that is is a good question - it could have to do with the machine construction or it might just be that Sennheiser has loosened the tolerances to reduce the number of rejects and save money. I have not confirmed this , but I remember the tightening jig was completely redesigned around the time of the AI introduction, the reason why was efficiency.
|
|
|
Post by matt@IAA on Sept 5, 2018 13:36:04 GMT -6
johneppstein If a person can be taught how to tap tune, a machine could do it too. The only reason they don't is because it is cost prohibitive. You could write an algorithm describing what a human does when they do this.
|
|
|
Post by bowie on Sept 5, 2018 18:01:06 GMT -6
There's this "expensive vs cheap" war that people seem to take a side on but it's not that simple. You really have to take it on a case by case basis. I've had more cheap gear fail but it's much easier to forget about those than it is the unit you paid $3,000 for. I've been shocked at some of the things I've found in some high end gear and it's sometimes being made by guys with highly regarded names in this biz. People you'd never think to question. Not just cheap parts here or there or sloppy work, but I've actually found mistakes in wiring where they got one channel correct and miswired something in the other. A $10,000 pair of custom mics where they didn't bother to even use the same brand of capacitors in the audio path of both mics. But, those are exceptions and I generally see not only smarter designs, but much better components used in high end gear. Most of the Chinese stuff I see is built quickly with the cheapest materials. Some rise above that though. And, others rise above it for a time, then slip in some area. Just like some of the high end builders do. I haven't noticed enough of a difference in reliability for it to influence my purchase the way that the sound quality will. I'll always buy for the sound, regardless of price, unless there's a serious issue going on with the company. Side note; a lot of people buy used high end gear (me included) and notice a higher instance of issues. Keep in mind that many people sell their gear rather than get it fixed or even looked at to see if there's a reason it's sounding bad. I unfortunately see this happen a lot. A LOT. I agree with all of that except for the eighth word from the end. Speaking as a person who is always looking for great deals on used gear and is somewhat technical, I think that you included a spurious "Un" on the word "Fortunately"... True, true. Half my collection of misfit gear is something that I got for a song and was able to repair. My favorite guitar is a Brazilian rosewood Bourgeois that I acquired outrageously cheap because it needed a few repairs. What I sadly see too often though is someone buying gear that someone else sold because it was acting up, with the issues themselves never being disclosed. It's the idea of facing several hundred dollars in repairs (or sometimes just maintenance) that the seller refuses to deal with so they see if they can pass the problem along to someone else.
|
|
|
Post by johneppstein on Sept 5, 2018 23:09:24 GMT -6
johneppstein If a person can be taught how to tap tune, a machine could do it too. The only reason they don't is because it is cost prohibitive. You could write an algorithm describing what a human does when they do this. With all respect, that's total nonsense. It's like saying that a machine could be taught to sing like Arertha Frankilin or Patsy Cline.
It's beyond the scope of what machines are capable of.
It's not "cost prohibitive", it's impossible because machines have no aesthetic sensibility. If it were true there would be no point in our continuing to make music or in pursuing any other art.
|
|
|
Post by matt@IAA on Sept 6, 2018 6:20:46 GMT -6
Haha. Machines can sing like Aretha. Queue up Spotify and hit play. It’ll do it all day, or sing like anyone else you want. Perfectly on pitch, loud or soft. A machine can make any pitch you like, but it won’t improvise, and it will only reflect the algorithm you give it.
But now you’re arguing for imperfection. I was talking about making products that have specifications and dimensions and tolerances.
|
|
|
Post by drbill on Sept 6, 2018 8:57:34 GMT -6
If it were true there would be no point in our continuing to make music or in pursuing any other art.
AI is going to change that John. Sad but true. Especially for a lot of non "high-end" applications of music. It's already here in it's infancy, and it will only get better, more efficient and more widespread. I'm so happy to have been a part of the generation I grew up in and to be transitioning OUT of this profession instead of transitioning IN. The future will look nothing like the past I'm afraid....
|
|
|
Post by johneppstein on Sept 6, 2018 10:35:14 GMT -6
If it were true there would be no point in our continuing to make music or in pursuing any other art.
AI is going to change that John. Sad but true. Especially for a lot of non "high-end" applications of music. It's already here in it's infancy, and it will only get better, more efficient and more widespread. I'm so happy to have been a part of the generation I grew up in and to be transitioning OUT of this profession instead of transitioning IN. The future will look nothing like the past I'm afraid.... Maybe. People have been predicting that for about 50 years, ever since the first "faux-Bach" fugues of the late '50s/early '60s. While "music-like" creations have happened there's nothing that can really fool somebody who knows anything about the music and I'm not holding my breath*. Maybe it will happen, but it hasn't yet despite all the "predictions". In fact really convincing AI hasn't happened in any field. And yes, I aware of the Japanese automatons that the AI researchers send out to give lectures and seminars - anyone who is really fooled by one of them is probably autistic.
I like science fistion as much as anyone (and probably more than most - you should see my book collection) but I don't confuse it with reality.
At any rate, it's not here yet.
* -The problem is that, being machine based, AI is not good at creating deviations from the program, at least not believable ones. Using Bach as an example, whaty made him such a good target for early AI is that many, if not most of his compositions were very grounded in the mathematics of music, which made them fairly easy to express in algorithms. The problem was that Bach himself would nearly always deviate to some degree from the mathematics at some point or points in the composition, which is what made his work interesting and where his genius lies. The AI composition programs could never really get that right, so, while they might fool a casual listener who has no formal music education on Bach's music it could never fool anyone who had actually studied his composition techniques. The same type of problem would face a programmer attempting to program a machine to tap tune the soundboard of a wooden instrument - it might be possible to get most of it done to the point of acceptability but it will have difficulty with any anomaly that falls outside the range of its algorithms. A human, OTOH, could conceivably turn those "weaknesses" into strengths. To put it more concisely, machines can't "think outside the box".
|
|
|
Post by chessparov on Sept 6, 2018 12:31:59 GMT -6
This all analogous to the development of chess programs. In my small way, I helped the developers of an early chess playing engine (Novag), by playing it game after game (about 80 games total IIRC). Typical result was 19 wins for me, with one draw at/near the end-because I got bored!
The majority of my older chessmaster friends, didn't have much faith in programs ever being good enough to beat the top players. How that has changed! Being younger helped convince me then, that the chess programs would eventually be successful in this.
Big +1 on my being happy to be born when I was (1959), and experiencing all the great music! Chris
|
|
|
Post by matt@IAA on Sept 6, 2018 12:40:20 GMT -6
Chess, interesting point. The most recent AlphaGo Zero was the first chess program not taught by humans. All older engines evaluated chess based on how humans play, given access to old games. In essence, they were limited to human application of the rules, and every play they made was an interpretation of a human play. AlphaGo Zero was given only the rules of chess and played itself. 24 hours later it went undefeated against the best version of Stockfish in 100 games (28 wins, 72 draws). Even more incredibly, the computer running Stockfish was capable of analyzing 70 million positions per second vs AlphaGo Zero at 80 thousand. In other words, computers are better at chess than we are now, not just as individuals but in the general sense. We can learn from AlphaGo.
|
|
|
Post by chessparov on Sept 6, 2018 12:55:59 GMT -6
Another "master" friend and I, estimate Alpha Zero's strength around 4000 Elo. It's an oversimplification, but some Grandmasters call playing computers like that...Playing alone against 50 Super-Grandmasters (who are conferring with each other) at once, in a single game!
Years before all this happened, I would have my private students, study the games of the greatest human players-Rather than concentrating on "their own" games. Similar concept.
Old (ironically) Chinese Expression..."From one thing, you can learn 10,000 things". This has more application in Pro Audio related things, than many realize. (so sayeth fortune cookie) Chris
|
|
|
Post by jin167 on Sept 7, 2018 1:57:50 GMT -6
A little of taste of high end boutique product for you gents
|
|