These clips are quite distinctive and I'm super excited to see the jams they're from! Hope y'all are having an alright Monday

Howdy. I think I've got the answer. I've written an audio fingerprinting algorithm to search the Phish archive (every song in "the spreadsheet"). Since it's cheating, I won't post the answer or try to claim the prize. But just for proof, I can post a hash (SHA-256) of the text of my answer (75b31641cc7977048378c8f47fee8b19c6473f15b9b8a905d65fea57d957f307). If anyone cares if I'm lying or thinks this is interesting, that can verify my answer later.

I've been experimenting with audio fingerprinting, so just built this system for kicks. I'm not sure if there's any real positive use for it - maybe identifying mislabeled tracks, etc... first time I'm trying out a newly released mystery jam.

I just ran the same audio fingerprinting algorithm and got "We Don't Talk about Bruno" - Encanto and "I Love Rocky Road" - Weird Al. Is that right @ucpete?

@jambot said:

Howdy. I think I've got the answer. I've written an audio fingerprinting algorithm to search the Phish archive (every song in "the spreadsheet").

I've been experimenting with audio fingerprinting, so just built this system for kicks. I'm not sure if there's any real positive use for it - maybe identifying mislabeled tracks, etc... first time I'm trying out a newly released mystery jam.

Hopefully it isn't tooo good, if you get what I'm saying .

Oh, and handle/post?

@jambot said:

Howdy. I think I've got the answer. I've written an audio fingerprinting algorithm to search the Phish archive (every song in "the spreadsheet"). Since it's cheating, I won't post the answer or try to claim the prize. But just for proof, I can post a hash (SHA-256) of the text of my answer (75b31641cc7977048378c8f47fee8b19c6473f15b9b8a905d65fea57d957f307). If anyone cares if I'm lying or thinks this is interesting, that can verify my answer later.

I've been experimenting with audio fingerprinting, so just built this system for kicks. I'm not sure if there's any real positive use for it - maybe identifying mislabeled tracks, etc... first time I'm trying out a newly released mystery jam.

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

@uctweezer said:

@jambot said:

Howdy. I think I've got the answer. I've written an audio fingerprinting algorithm to search the Phish archive (every song in "the spreadsheet"). Since it's cheating, I won't post the answer or try to claim the prize. But just for proof, I can post a hash (SHA-256) of the text of my answer (75b31641cc7977048378c8f47fee8b19c6473f15b9b8a905d65fea57d957f307). If anyone cares if I'm lying or thinks this is interesting, that can verify my answer later.

I've been experimenting with audio fingerprinting, so just built this system for kicks. I'm not sure if there's any real positive use for it - maybe identifying mislabeled tracks, etc... first time I'm trying out a newly released mystery jam.

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

I wish I had any sort of clue of what you all are talking about . . . It sounds smart . . . I don't think I'm smart enough to follow this. Is "kang" related to "Kung?"

@uctweezer said:

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

PM'ing my answer. Yeah, I'm not sure, but it seems pretty robust. I tried comparing to the MJM history spreadsheet, it seems to get about 95% correct on first guess, with some caveats - only running on the "one-song" samples (not attempting to split into multiple clips - today's I split into two files by hand), skipping the non-phish, etc. I'd like to go back and see what it is getting wrong and why. Some annoyances with matching song names (eg the spreadsheet will have names like "__ > Jam" , etc).

Yeah, I don't want to spoil the fun here or take the prize and don't plan to put online, but since I recently decided to store all the shows I could get locally, I thought I'd give it a try.

@A_Buddhist_Prodigy said:

@uctweezer said: [quote @jambot said: [quote]Is "kang" related to "Kung?"

No. Kang is related to Kodos.

Wait, this isn't the Simpsons thread?

@jambot said:

@uctweezer said:

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

PM'ing my answer. Yeah, I'm not sure, but it seems pretty robust. I tried comparing to the MJM history spreadsheet, it seems to get about 95% correct on first guess, with some caveats - only running on the "one-song" samples (not attempting to split into multiple clips - today's I split into two files by hand), skipping the non-phish, etc. I'd like to go back and see what it is getting wrong and why. Some annoyances with matching song names (eg the spreadsheet will have names like "__ > Jam" , etc).

Yeah, I don't want to spoil the fun here or take the prize and don't plan to put online, but since I recently decided to store all the shows I could get locally, I thought I'd give it a try.

5% is also about how often I spruce up the tapes I’m dealing with, whether making a SBD / AUD matrix or adjusting an old tape to the right speed (the latter of which I’d expect to fool your tool every time)

@uctweezer said:

@jambot said:

@uctweezer said:

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

PM'ing my answer. Yeah, I'm not sure, but it seems pretty robust. I tried comparing to the MJM history spreadsheet, it seems to get about 95% correct on first guess, with some caveats - only running on the "one-song" samples (not attempting to split into multiple clips - today's I split into two files by hand), skipping the non-phish, etc. I'd like to go back and see what it is getting wrong and why. Some annoyances with matching song names (eg the spreadsheet will have names like "__ > Jam" , etc).

Yeah, I don't want to spoil the fun here or take the prize and don't plan to put online, but since I recently decided to store all the shows I could get locally, I thought I'd give it a try.

5% is also about how often I spruce up the tapes I’m dealing with, whether making a SBD / AUD matrix or adjusting an old tape to the right speed (the latter of which I’d expect to fool your tool every time)

Additionally, the string of MJMs that I made starting in 2018 and hosted in 18-19 and sporadically since did not use spreadsheet AUDs or SBDs for the most part so those likely have mixed results if trawling for matching hashes. I’m curious about your tool’s performance on those.

This is super cool that you did this, and even cooler that you’re being a good sport by not sharing the answers.

@wforwumbo said:

@uctweezer said:

@jambot said:

@uctweezer said:

Cool! We've been talking about this for years, and finally someone did it

I wonder how source dependent it is - I didn't use the spreadsheet source. We had even considered introducing some (imperceptible) noise to fuck with people using fingerprinting schemes, but decided no one would ever spend that much time cheating for a $10 code haha

PM me if you want to kang your answer (I didn't get a matching hash but only tried a couple date formats)

PM'ing my answer. Yeah, I'm not sure, but it seems pretty robust. I tried comparing to the MJM history spreadsheet, it seems to get about 95% correct on first guess, with some caveats - only running on the "one-song" samples (not attempting to split into multiple clips - today's I split into two files by hand), skipping the non-phish, etc. I'd like to go back and see what it is getting wrong and why. Some annoyances with matching song names (eg the spreadsheet will have names like "__ > Jam" , etc).

Yeah, I don't want to spoil the fun here or take the prize and don't plan to put online, but since I recently decided to store all the shows I could get locally, I thought I'd give it a try.

5% is also about how often I spruce up the tapes I’m dealing with, whether making a SBD / AUD matrix or adjusting an old tape to the right speed (the latter of which I’d expect to fool your tool every time)

Additionally, the string of MJMs that I made starting in 2018 and hosted in 18-19 and sporadically since did not use spreadsheet AUDs or SBDs for the most part so those likely have mixed results if trawling for matching hashes. I’m curious about your tool’s performance on those.

This is super cool that you did this, and even cooler that you’re being a good sport by not sharing the answers.

Yeah, I think changing the speed would definitely mess it up. I think different mp3 encodings of same recording would handle well. Different recordings of the same show, I'm not sure. When I get a chance to look at the details, I'll let you know.

I always upload FLACs to SoundCloud (and have for a couple years since they enabled lossless audio) so I think you're golden on the encoding robustness. Also, I almost never use the same source as the spreadsheet (and use a ton of SBDs, which aren't on the spreadsheet), so it seems you're golden on different sources as well. Would be cool to characterize the failures - I can likely tell you why they failed, given a list...

Also it would be fantastic to get the track timings for all past MJMs - I have many of them in text files but the data are a mess and not necessarily accurate (for reasons I won't elaborate on here only because it's not that interesting lol), and have long thought about crowdsourcing an effort to annotate them all ... but your tool would make this trivial and I could just add a column to the results spreadsheet, linking back to phish.in (which began with the set of spreadsheet sources but has diverged a bit over the years).

Really cool stuff, thanks for sharing

7/9/94 Antelope
7/9/98 Tweezer

you guys really gonna let the blog win?

does the show sound outdoors or indoors to you? does it sound like a large crowd or a small crowd? does it sound like the recording was made up front or way in the back?

I don't know. All I can hear is the chicks in the front row.