[10:39.990 --> 10:42.290] ...or if we miss, like, 40% of them. [10:42.510 --> 10:51.610] I don't think it's actually that much in practice, but we just need to find enough fragments from this text data set that we have an arsenal of candidates. [10:53.410 --> 11:06.470] And it is also fun, as you go through songs, accumulating this list of overrides or this list of, like, words that weren't in the base Carnegie Mellon pronouncing dictionary that we are filling in as we go. [11:06.470 --> 11:10.790] You might be able to recognize some familiar songs among these blocks. [11:11.190 --> 11:13.970] We've got Oasis, Wonderwall here. [11:14.350 --> 11:16.690] We've got We Didn't Start the Fire in this block. [11:17.030 --> 11:20.390] We've got a large section devoted to Family Affair by Mary J. Blige. [11:21.290 --> 11:25.470] There's some Bohemian Rhapsody over here. [11:25.770 --> 11:32.910] And, you know, it's kind of a scrapbook of different unusual pronunciations that I've encountered over the course of this project. [11:34.110 --> 11:35.390] Let's do another demo. [11:35.630 --> 11:42.670] And for this one, I would like to intentionally choose a song that has a second part. [11:43.470 --> 11:51.570] Total Eclipse of the Heart is a karaoke classic that starts with Turn Around, which is sung by a chorus of... [11:51.570 --> 11:56.850] I think in the video it's like a chorus of ghost children, but you will be the chorus of ghost children here. [11:56.850 --> 12:01.770] And let's go... let's go to our visualizer here. [12:04.300 --> 12:06.230] And choose Total Eclipse. [12:07.950 --> 12:18.310] And for the replacements, I guess I'd like to invite you to take your pick of three that I thought might be appropriate for hope. [12:18.310 --> 12:23.910] We have the Panama Papers, a list of company names in the Panama Papers release. [12:24.850 --> 12:30.670] Today, I went and got, from the Internet Archive, I got Free Software, Free Society. [12:31.290 --> 12:40.530] The open... the free software kind of ur-text, or one of the collections of free software. [12:41.350 --> 12:44.390] And then there's a third one, Hackers. [12:44.530 --> 12:46.350] Yeah, I heard that there was a screening of Hackers. [12:46.510 --> 12:50.750] So we have Hackers Dialogue and Hackers Stage Directions as two sources. [12:51.070 --> 12:54.870] Does anyone have one of those that especially speaks to them that they want to shout out? [12:55.610 --> 13:00.550] All right, we're doing Hackers Stage Directions to the tune of Total Eclipse of the Heart. [13:02.050 --> 13:05.810] And again, yeah, you will be the first singer here. [13:05.810 --> 13:10.270] You will sing The Small Crowd to the tune of Turn Around, and I will sing the rest of it. [13:10.910 --> 13:12.230] Okay, here we go. [13:19.970 --> 13:25.550] Turn Around, Tate stands alone in the hall watching the tune closely. [13:25.550 --> 13:27.570] He'd rather not do this now. [13:28.010 --> 13:35.810] Turn Around, Tate studies and reverse engineers the garbage while a woman is seated at the desk. [13:36.010 --> 13:43.310] Turn Around, Serial emerges from the desk between her and writes the video, change your diagram lights. [13:43.410 --> 13:45.050] Turn Around, Tate, talk, and жизнь. [13:45.050 --> 13:50.870] He sees his computer Lucy being carried away The Gibson dies in a flash of light. [13:51.090 --> 13:53.950] Turn Around, Tate, talk, and love. [13:54.290 --> 13:56.790] Serial takes off to fix the phone. [13:57.890 --> 14:00.290] Turn Around, Tate's alone. [14:00.290 --> 14:02.470] Serial takes off to fix the phone. [14:03.230 --> 14:08.910] Turn Around, Tate, talk, and reverse engineers the garbage while the Da Vinci Viruses dying. [14:10.790 --> 14:14.390] Like one only notices and takes off the helmet. [14:14.390 --> 14:17.270] Only Joey needs to put his hands up. [14:40.830 --> 14:46.810] If it's you only find the test of a police type vehicle. [14:51.060 --> 14:54.920] But there's a man waits for some time. [14:55.660 --> 14:57.220] Freaks us two-handed. [14:58.040 --> 15:02.360] A fly-up that gives us in a flash of light. [15:02.360 --> 15:04.900] Lauren Murphy's records. [15:05.480 --> 15:08.180] They walked away on Gary's umbrella smoke and [15:14.060 --> 15:14.360] dry. [15:14.760 --> 15:17.980] A deep sight in the observational appearance. [15:18.360 --> 15:21.540] She indicates a pile of college applications. [15:22.700 --> 15:24.760] I don't believe I'm fine. [15:25.580 --> 15:28.200] They live from a wasteful sunshine. [15:29.160 --> 15:32.860] A red man on the glass to clean and cereal. [15:33.220 --> 15:34.640] Acknowledge each other. [15:34.640 --> 15:37.880] They have darkly exchanged mindsets. [15:40.360 --> 15:42.460] A sticky lock ensues. [15:42.680 --> 15:45.060] A full leopard skin muscle shark. [15:47.580 --> 15:48.760] Oh, thank you. [15:51.840 --> 15:55.420] The motor circle speeds up into the night. [15:56.040 --> 15:57.960] A fast is only a wrap number. [15:59.660 --> 16:01.240] Let's see next this way. [16:01.600 --> 16:04.500] Serena lies under the desk. [16:05.560 --> 16:07.400] Oh, we can skip this instrument. [16:09.480 --> 16:10.440] There it is. [16:12.180 --> 16:13.060] I didn't come back. [16:13.160 --> 16:14.200] You are the first line back. [16:18.320 --> 16:22.420] I can't believe your sow, let's see. [16:22.720 --> 16:23.080] Let's see. [16:23.080 --> 16:24.640] I don't believe I know you are. [16:24.960 --> 16:26.720] Let's see. [16:35.140 --> 16:37.880] I don't believe you will. [16:37.880 --> 16:38.460] I don't believe I know you are. [16:38.460 --> 16:38.580] I don't think I'm okay. [16:39.600 --> 16:40.200] egyę. [16:40.200 --> 16:42.840] The UI, a friend, no status. [16:43.500 --> 16:46.500] The kids sitting down in a flash of light. [16:47.420 --> 16:49.100] Now, day really looks lost. [16:49.520 --> 16:52.700] And loses what his horror is about to come up. [16:53.100 --> 16:56.480] Street vendors have computer birds and food and software. [16:58.340 --> 17:02.220] The fiber-playing rule, the fashionable part. [17:02.560 --> 17:05.640] And seizure which you got, the secret service. [17:05.640 --> 17:05.700] Yes! [17:08.020 --> 17:10.500] He'd rather not do this now! [17:14.560 --> 17:17.460] A red mother that was soon. [17:17.600 --> 17:20.420] Date and serial acknowledge each other. [17:20.920 --> 17:23.240] It has the logo of Stantina. [17:24.800 --> 17:28.800] Exterior view, the virus repeats its demand. [17:35.480 --> 17:39.200] The motorcycle speeds up into the night. [17:39.480 --> 17:41.720] Late night is horror watching TV. [17:43.440 --> 17:45.000] The system's displayed. [17:45.500 --> 17:47.540] A full-legged skin muscle shirt. [17:52.290 --> 17:54.850] Full-legged skin muscle shirt. [17:54.850 --> 17:56.190] The [18:01.780 --> 18:14.300] concept is so loud. [18:18.300 --> 18:19.200] That's it! [18:19.600 --> 18:20.820] Thank you so much! [18:21.900 --> 18:22.620] Okay! [18:23.040 --> 18:24.240] That's the premise. [18:25.160 --> 18:36.420] Before we get to any more demos, and keep in mind that you can request any combination of sources, We have sources and texts, source texts and songs in the catalog via that QR code when it comes up. [18:37.160 --> 18:40.140] But I want to talk a little bit about where does this data come from? [18:40.560 --> 18:54.980] And I think this is actually, to me, the heart of the project is finding out what is accessible, like what can you find in the world or on the web or from each other that's interesting to revisit through this kind of karaoke kaleidoscope lens. [18:55.200 --> 19:00.100] So I just want to spotlight a few of the data sets, all of which I think are available to request. [19:00.180 --> 19:01.160] If you want to. [19:02.180 --> 19:10.800] One place is just getting data in the wild on the Internet using scrapers or just copying and pasting text yourself. [19:11.180 --> 19:16.380] There's a number of Yelp review data sets in the source text catalog that you can choose. [19:16.640 --> 19:23.420] And some of those are from, I think maybe 10 years ago, it seemed easier to scrape data from Yelp than it does now. [19:23.700 --> 19:29.100] So typically when I think of something I want to get from Yelp now, I'll just go and copy paste a few pages of it. [19:29.100 --> 19:34.760] That is actually simpler than trying to figure out how to get around the modern scraping protocols. [19:35.320 --> 19:37.820] But yeah, the Internet is a good place to find data. [19:38.660 --> 19:47.060] Sometimes there are sites on the Internet like Kaggle that really specialize in well-structured data that you don't have to do any of the structuring yourself. [19:47.060 --> 19:57.540] So with the Yelp, you might need to sort it by star rating and split it into sentences or just getting the data in the first place is hard. [19:57.860 --> 20:08.780] On Kaggle, which is a site for data scientists and often hosts competitions where a company will open-source a subset of their data and invite people to do various challenges. [20:09.160 --> 20:16.500] Like often it's predicting what movie or song or whatever someone will like based on a history of liked movies and songs. [20:16.740 --> 20:24.540] They'll do that because they're interested in like leveraging the open-source community's smarts to solve one of their company problems. [20:24.880 --> 20:32.120] But incidentally, they will publish a bunch of karaoke compatible fragments of text data that we can repurpose. [20:32.120 --> 20:38.200] So Quora questions are one of my favorite in the catalog if you were wondering what to request. [20:38.440 --> 20:42.480] And one more note about requests, you can request something for the crowd to sing. [20:42.600 --> 20:43.900] You don't have to do it yourself. [20:44.180 --> 20:50.720] So if you want to say like this is a group sing or request it as question mark question mark, that's cool too. [20:51.780 --> 20:53.880] You can source data from the U.S. government. [20:54.480 --> 20:56.880] One of my favorite data sets in the catalog. [20:56.880 --> 21:02.060] Actually, I'll show you this after this review is Form 1040. [21:02.300 --> 21:05.100] And this one we use as a visual text replacement. [21:05.720 --> 21:06.520] I'll show you how that works. [21:06.960 --> 21:08.920] But yeah, the government is a good source of data. [21:09.660 --> 21:11.040] Wikis are great, obviously. [21:11.240 --> 21:14.440] People are contributing text to a shared knowledge base. [21:14.580 --> 21:16.140] It's very in the spirit of the show. [21:16.140 --> 21:22.460] So going to Wikipedia or the Bellatro Wiki to name the two most prominent wikis in my life. [21:23.960 --> 21:29.280] They're good places to grab some examples to replace things with. [21:29.900 --> 21:32.620] One of my favorite data sets in the catalog is this one. [21:32.900 --> 21:37.140] Banner Depot 2000, which Yu Feng Zhao and Richard Huang curate. [21:37.980 --> 21:40.620] They are two data artists based in Seattle and New York. [21:40.620 --> 21:45.680] And this is a set of Internet banner ads from the early 2000s and late 90s. [21:45.920 --> 21:47.740] It's got about 20,000 of these. [21:48.020 --> 21:51.480] And the great thing is we have both the text and the images. [21:51.660 --> 21:56.900] So we can play back a karaoke song and sing directly from the banner ad. [21:56.980 --> 21:59.440] We don't need to use the text in isolation. [21:59.440 --> 22:01.520] We can, like, sing the text in context. [22:01.720 --> 22:06.620] So if you want to see what that looks like, request Banner Depot or banner ads from the early Internet. [22:07.780 --> 22:15.140] Something we've done before, but I don't think we have time to do tonight, is to do a crowdsource data set. [22:15.220 --> 22:21.260] Where we publish a form and everyone in the crowd submits, in this case, their biggest fears. [22:21.640 --> 22:22.980] Like, this was for a Halloween show. [22:23.260 --> 22:25.460] We crowdsourced a bunch of fears. [22:26.220 --> 22:30.280] And that is a data set you can select that crowdsourced fears from last year. [22:31.380 --> 22:32.620] Yeah, that's another option. [22:33.240 --> 22:34.320] We can get it from each other. [22:34.320 --> 22:37.840] And then this is maybe a subset of the last two. [22:38.960 --> 22:43.840] LARPA, which is this artist makerspace that I am a member of in Brooklyn. [22:44.200 --> 22:45.580] We run public events. [22:45.720 --> 22:49.880] One of the most relevant to this project from the last year was Small Batch. [22:50.040 --> 22:51.900] Which was a data set farmer's market. [22:52.120 --> 22:54.860] Where we invited people to bring small data sets. [22:54.960 --> 22:55.600] Which could be anything. [22:55.600 --> 22:59.180] Any structured collection of images or text. [22:59.460 --> 23:04.260] From your life or from a corner of the world or Internet that you have trolled. [23:05.060 --> 23:08.160] And I think that in itself is interesting. [23:08.380 --> 23:09.700] That's kind of the message of this event. [23:09.860 --> 23:11.020] And maybe also of this show. [23:11.220 --> 23:15.060] Is like, you add a lot just by noticing things and collecting them. [23:15.240 --> 23:18.480] And that is all it really takes to make a data set. [23:19.440 --> 23:22.140] It's not something that's fully abstracted from us. [23:22.260 --> 23:24.540] That only lives in like back-end servers. [23:25.300 --> 23:26.440] But this crowd knows that. [23:27.200 --> 23:28.800] Some frontiers that I'm excited about. [23:28.920 --> 23:30.980] That have not worked their way fully into the system yet. [23:31.320 --> 23:32.180] For sourcing data. [23:32.380 --> 23:34.560] I think public bulletin boards are going to be great. [23:35.100 --> 23:36.480] They're dense with text. [23:36.620 --> 23:38.080] They have like a visual character. [23:38.080 --> 23:41.060] I'd like to do a kind of similar thing to the forum 1040. [23:41.760 --> 23:42.520] With bulletin boards. [23:42.740 --> 23:46.060] FOIA requests, obviously, we need to incorporate into the system. [23:46.780 --> 23:52.640] You can, using the Freedom of Information Act, request government communications of any sort. [23:53.080 --> 23:58.180] I think it would be particularly interesting to see how government employees are using chatbots. [23:58.700 --> 24:03.660] And FOIA requests a collection of chat logs to use for karaoke purposes. [24:05.060 --> 24:07.580] Radio scanners inspired by the earlier talk. [24:08.060 --> 24:09.940] This is a slide I made this afternoon. [24:10.180 --> 24:13.220] We should use radio scanners as a source of karaoke text. [24:13.660 --> 24:14.520] Focus groups. [24:15.040 --> 24:16.160] Live transcripts of the show. [24:16.400 --> 24:20.040] These are all up and coming frontiers in karaoke data acquisition. [24:22.680 --> 24:29.720] I'll just do a whirlwind tour of some of the way that the features of this show have evolved over the last year or two. [24:31.120 --> 24:35.820] At first, in version one, this was running in the terminal. [24:35.980 --> 24:40.680] This was just the natural language processing stuff happens in complete isolation from the song. [24:40.840 --> 24:44.240] And the way the show would work was get a terminal window up. [24:44.620 --> 24:46.560] Get a browser window up. [24:46.760 --> 24:51.080] Start playing the song and then press enter every time I wanted to advance a line. [24:51.580 --> 24:54.500] And that really was making karaoke harder. [24:54.500 --> 25:01.100] I suppose we've strayed from our core mission, in a sense, if we've made karaoke easier on the host. [25:01.540 --> 25:04.100] It's gotten harder in different ways for the audience. [25:05.680 --> 25:10.920] Version two, we moved the karaoke into the browser, but still... [25:10.920 --> 25:17.060] Or moved the timing away from having to press enter, but it was still two windows. [25:17.060 --> 25:22.700] Then we moved the audio into the browser and had syllable level timestamps. [25:22.820 --> 25:30.880] And we're really moving into the modern age here with a persistent header that showed you what song and text was playing at any given time. [25:31.340 --> 25:36.620] And that brings us up to the present day with version four, where we have the cue in the side rail. [25:36.940 --> 25:40.280] You can select a song and a text from these menus at the top. [25:40.280 --> 25:41.520] There's a title card. [25:41.720 --> 25:42.380] It's really... [25:42.380 --> 25:46.040] I mean, this is now ready for the world. [25:46.300 --> 25:48.120] You can click on the song to navigate. [25:48.760 --> 25:50.160] Those are convenience features. [25:52.340 --> 25:53.440] How do we run the show? [25:54.220 --> 25:55.400] There is a live show. [25:55.640 --> 26:03.440] There's a regular one at Wonderville every two months that we often use as like a feature testing ground. [26:05.060 --> 26:05.900] This is... [26:06.440 --> 26:07.440] This is Wonderville. [26:07.620 --> 26:08.080] It's a... [26:08.080 --> 26:10.120] If you don't know about Wonderville, I should tell you about it. [26:10.280 --> 26:16.260] This is an excellent space in Brooklyn on the border of Bed-Stuy and Bushwick. [26:16.480 --> 26:21.420] It is essentially a rotating museum of DIY arcade cabinets. [26:21.780 --> 26:23.240] And it's free to the public. [26:23.480 --> 26:25.100] It's a bar as well. [26:25.220 --> 26:28.040] So it's a bar attached to a DIY arcade. [26:28.760 --> 26:32.760] And they have shows, including robot karaoke every other month. [26:32.760 --> 26:36.600] Here's this QR code that you can use to sign up again. [26:36.940 --> 26:38.420] That's the last time I pointed out. [26:39.640 --> 26:44.200] And if you do follow that QR code, you will see this list of songs. [26:46.500 --> 26:48.020] This list of texts. [26:48.320 --> 26:51.080] And then the option to enter your name or request a group sing. [26:52.800 --> 26:55.920] There are some upcoming shows for robot karaoke. [26:56.180 --> 27:00.400] Not just at Wonderville, but also at Game Dev Stalwart's Boshi's Place. [27:00.400 --> 27:02.820] I don't know if there are any Boshi's Place heads in the crowd. [27:03.400 --> 27:12.740] And then I'm going to San Francisco and Seattle later this month for some shows at PAX and at Gray Area in San Francisco. [27:13.360 --> 27:13.600] Okay. [27:14.120 --> 27:19.940] Demo 3+, this is where I hope you come in and we'll have at least some suggestions that we can sing through. [27:20.340 --> 27:24.060] If not, you're also welcome to come up and ask questions. [27:25.600 --> 27:29.940] So I'll end the walk through there and say thanks, but we have a show to get to. [27:30.220 --> 27:31.460] So thank you. [27:39.600 --> 27:40.400] And refreshing... [27:41.300 --> 27:43.100] We don't have any new requests. [27:45.220 --> 27:45.620] So... [27:45.620 --> 27:47.480] Oh, we have someone walking up to a mic. [27:47.680 --> 27:48.980] Is this a question or a request? [27:49.020 --> 27:49.920] It's a question. [27:50.240 --> 27:50.700] Yeah, go for it. [27:50.760 --> 27:52.880] Just curious, how do you do the syllable calculation? [27:52.880 --> 27:54.160] In order to... [27:54.160 --> 27:59.500] So the syllable counts come from that resource that I mentioned, the pronouncing dictionary. [27:59.940 --> 28:00.240] So... [28:01.180 --> 28:02.380] CMU pronouncing dictionary. [28:02.520 --> 28:04.680] Let me see if I can find the intro slide here. [28:05.320 --> 28:05.920] Each of these... [28:06.580 --> 28:08.560] I should have broken this down in more detail. [28:08.860 --> 28:13.280] Like, each of these words is paired with a set of phones or phonemes. [28:14.100 --> 28:16.780] And some of them are consonants and some of them are vowels. [28:17.240 --> 28:20.340] The count of syllables is just the count of vowels in the word. [28:20.560 --> 28:23.160] And you can't have words where it's, like, ambiguous. [28:23.340 --> 28:25.780] Whether it's one syllable or two. [28:25.980 --> 28:27.460] So it might be, like, fire. [28:27.720 --> 28:30.540] You can say it as, like, one, fire. [28:30.720 --> 28:32.380] Or you can say it as fire. [28:33.240 --> 28:34.120] And that... [28:34.120 --> 28:35.660] It's pretty subjective. [28:35.980 --> 28:38.120] You might say, in this song it's working as two syllables. [28:38.360 --> 28:39.700] So I'll treat it as two. [28:40.680 --> 28:45.200] But usually the CMU's first pronunciation is the most dominant one. [28:45.380 --> 28:46.640] Or the most prevalent one. [28:47.120 --> 28:49.940] So it's a pretty good first pass. [28:51.640 --> 28:52.720] Yeah, thank you. [28:55.380 --> 28:56.540] Yeah, in the... [28:56.540 --> 28:57.320] This is amazing. [28:57.620 --> 28:57.940] Thank you. [28:58.540 --> 29:08.040] In the text that you're using to replace the lyrics with, what do you have to do to make phrases or sentences? [29:08.040 --> 29:13.620] So the question is, what do you have to do to make phrases or sentences out of the text? [29:14.600 --> 29:18.580] I think it's more like we're cutting down a text that starts... [29:18.580 --> 29:21.840] You take the full text and you break it up into phrases. [29:23.340 --> 29:30.380] Like, we're not doing much assembly from the bottom up of saying, like, how would these two fragments fit together into a line? [29:30.520 --> 29:35.760] Each line that you see, by and large, is a full phrase from the dataset directly. [29:37.480 --> 29:38.540] So we're not... [29:38.540 --> 29:40.180] Okay, but is it like a logical phrase? [29:40.320 --> 29:41.140] I guess that's my question. [29:41.160 --> 29:43.140] Or is it just going by the sentence period? [29:43.140 --> 29:44.000] Or is it... [29:45.020 --> 29:47.620] Seeking within the sentence and taking only part of it? [29:48.520 --> 29:51.160] In this version, there's no... [29:51.160 --> 29:54.100] There's no attempt to make the lines flow into each other. [29:54.500 --> 29:55.760] Or to make them... [29:56.660 --> 29:59.220] Make the pieces make logical sense next to each other. [29:59.220 --> 29:59.440] Yeah. [30:00.380 --> 30:01.420] There is some... [30:02.160 --> 30:11.940] In processing the text on the first pass, like when we're getting the set of fragments from the text, there is some, like, detection of noun phrases. [30:11.940 --> 30:21.580] So you can use a library that says, treat each noun phrase in addition to each sentence as a candidate fragment. [30:21.580 --> 30:22.040] Okay. [30:22.300 --> 30:23.680] And so there's some of that. [30:24.100 --> 30:29.560] But most of the time, it's good enough to just take the sentence breaks as they exist. [30:30.260 --> 30:32.380] And you don't need to go beyond that. [30:32.480 --> 30:33.420] I think for some... [30:33.420 --> 30:42.800] For some books, for example, where the sentences are very long, and they stand no chance of being short enough to fit in a line, it's important to do that noun chunk extraction. [30:43.220 --> 30:50.420] And so that's kind of like a heavy artillery that I'll activate when I notice that there's just like not enough short sentences in the first pass. [30:50.420 --> 30:51.020] Thank you so much. [30:52.180 --> 30:52.660] Yes. [30:54.040 --> 30:56.520] I don't know which one of you was first. [30:56.680 --> 30:57.080] Let's go here. [30:58.000 --> 30:59.820] Is this something we can run on our own? [30:59.960 --> 31:00.640] And if so, how? [31:01.100 --> 31:11.820] So the combination, the song and text combination, is not yet publicly available, but I'm trying to figure out a way to do it that doesn't immediately get taken down. [31:12.700 --> 31:27.880] I think that right now the most promising thing to me seems to be open sourcing the data standard that is like the way the song text and the replacement lyrics connect to each other, and then also open sourcing as many of the data sets as I can. [31:28.780 --> 31:45.340] I think the song part, even though there's a strong case in my mind that this is transformative reuse and completely legal if you had the resources to argue it, I think that there's a part of me that's wary of putting up song lyrics in any form on a publicly hosted site. [31:45.340 --> 31:47.260] So it's something I'm thinking through. [31:47.620 --> 32:00.280] If you have thoughts about like how to approach that issue of like releasing a copyright proof or like a copyright, you know, common sense version of this, talk to me after. [32:00.860 --> 32:01.500] Thank you. [32:03.400 --> 32:04.360] Okay, great. [32:04.580 --> 32:05.360] Two words with one stone. [32:06.240 --> 32:07.340] Yes, go for it. [32:09.780 --> 32:10.600] Oh, yeah, sorry. [32:10.760 --> 32:12.200] Could you go to the mic? [32:12.420 --> 32:12.860] Oh, sorry. [32:13.000 --> 32:13.060] Good. [32:13.200 --> 32:13.360] Yes. [32:13.880 --> 32:14.380] Oh, thank you. [32:15.160 --> 32:15.520] Awesome. [32:16.420 --> 32:16.940] Thank you, sir. [32:18.000 --> 32:20.280] What's stopping you from using the resources we already have? [32:20.400 --> 32:22.760] We have lawyers, and we have the EFF right here. [32:24.040 --> 32:25.340] I think it's just this... [32:26.260 --> 32:27.580] There shouldn't be anything stopping you. [32:27.720 --> 32:28.580] Not having talked to them yet. [32:28.820 --> 32:29.260] Oh, okay. [32:29.580 --> 32:31.000] I'll do anything I can to help you. [32:31.240 --> 32:32.240] Hi, my name is Pam. [32:32.240 --> 32:37.640] I'm the daughter of an electrical engineer, and an honor scholar myself from Rutgers. [32:38.120 --> 32:38.400] Amazing. [32:38.720 --> 32:39.680] I can do almost anything. [32:41.000 --> 32:41.800] Thank you. [32:43.100 --> 32:43.800] Rock on. [32:45.660 --> 32:50.240] Any interest in extending into translingual sorts of songs? [32:50.560 --> 32:56.580] Because I can bet a lot of people would love to sing Gangnam style mixed up into English. [32:56.580 --> 32:58.040] Oh, we've already done. [32:58.100 --> 33:03.540] We do, in fact, have some Mandarin songs and some Mandarin text data sets. [33:03.640 --> 33:10.300] I don't know if anyone speaks Chinese, but we have simplified Chinese. [33:10.700 --> 33:13.460] Or actually, I don't speak Chinese, so I do not know. [33:13.700 --> 33:24.420] But one convenient thing about this particular transliteration is you don't need a pronouncing dictionary to count the syllables in a language where each character is one syllable. [33:24.420 --> 33:30.220] So we have done this with crowds who speak Chinese. [33:30.720 --> 33:33.700] And they've said it works well enough. [33:33.920 --> 33:41.080] Like, we don't get the high quality of rhymes that we would in English because we're just matching each character and treating that as a syllable. [33:41.440 --> 33:43.460] I think that there are also emerging... [33:43.460 --> 33:49.680] Increasingly, there are resources for pronouncing different languages emerging that are as stable as the CMU. [33:49.680 --> 33:53.220] I just found one this week for Spanish that I'm very interested in trying out. [33:54.120 --> 33:58.040] But yeah, very, very much there is interest in translingual. [33:58.340 --> 34:00.280] We can try one if you request it. [34:00.920 --> 34:01.700] I don't do Chinese. [34:01.940 --> 34:02.980] I can do Japanese, though. [34:03.400 --> 34:03.400] Okay. [34:03.400 --> 34:04.220] We can... I mean, great. [34:04.440 --> 34:04.620] Okay. [34:08.970 --> 34:09.490] Hey. [34:10.250 --> 34:10.570] Hello. [34:11.310 --> 34:11.590] Hi. [34:11.870 --> 34:12.770] Great live demo. [34:13.050 --> 34:17.390] I was wondering if you could talk about how you choose songs and data sets. [34:17.990 --> 34:18.590] Sure. [34:18.990 --> 34:28.430] I think, for songs, something I've noticed is that it helps when each verse has the same metric pattern. [34:28.770 --> 34:37.290] Because, say, like, to take a counterexample, Bob Dylan is not a good artist for doing these kinds of replacements in general. [34:37.290 --> 34:40.490] Like, there are some songs, maybe, from early Bob Dylan. [34:41.450 --> 34:50.630] But mostly he's sort of like rapid fire coming at you with different variants on phrases and different patterns that you might not encode if you're listening to the song. [34:50.710 --> 34:54.830] You might just remember it as the story or you remember it as, like, the feel of the song. [34:55.050 --> 34:59.470] I think that a lot of... a lot of rap is that way, too. [34:59.810 --> 35:02.170] Like, maybe there's also an era of rap. [35:02.170 --> 35:04.450] Like, Tribe Called Quest might work. [35:04.710 --> 35:09.110] I think Kendrick Lamar would be really difficult because they... [35:09.110 --> 35:14.510] Like, all Kendrick songs are playing with meter from verse to verse. [35:14.710 --> 35:22.630] So I think stable and memorable lines that are sort of earworms are a good recipe for a song. [35:23.230 --> 35:28.110] And I think for text, I really love reviews. [35:28.390 --> 35:33.250] And I've often thought about, like, what makes reviews such good data sets. [35:33.370 --> 35:34.270] We haven't done any yet. [35:34.390 --> 35:38.850] But we should do negative Glassdoor reviews, I would propose, if we have time for another song. [35:39.530 --> 35:43.070] And I think one reason that that works is, like, they're so... [35:43.830 --> 35:48.990] They're in a format that's trying to get as much out as possible in a compact space. [35:50.450 --> 35:51.710] And they're also emotional. [35:51.990 --> 36:04.150] So if you pair, like, an emotional data set, like, negative reviews of places that people are forced to spend all of their time, with a song that's about, like, a breakup or a bad relationship, it often goes really well. [36:04.270 --> 36:06.110] Because there's, like, that alignment. [36:06.410 --> 36:07.590] So there's also a kind of... [36:07.590 --> 36:13.650] Aside from choosing ones that are good on their own, there's also, like, a sort of pairing of text and song. [36:13.650 --> 36:19.150] Like, we need to hire a sommelier to take this art form to the next level. [36:20.030 --> 36:20.810] Good question. [36:20.930 --> 36:21.170] Thank you. [36:22.110 --> 36:23.990] So I'm going to be drunk with power for a second. [36:23.990 --> 36:26.310] And I'm going to abuse this to ask my question. [36:26.630 --> 36:27.090] Go for it. [36:27.310 --> 36:35.650] Weird Al is actually really well known for reaching out to the artists that he parodies and gets their buy-in before actually publishing work. [36:36.650 --> 36:41.850] Have you managed to reach out to any artists and gotten their thoughts on this? [36:41.950 --> 36:44.030] Because I'm very curious to know what they would think of this. [36:44.990 --> 36:46.030] That's a great question. [36:46.250 --> 36:49.210] I haven't talked to any of the artists on this list. [36:49.630 --> 36:50.390] Maybe that's... [36:51.330 --> 36:56.190] I don't think that I had the opportunity and declined it. [37:00.840 --> 37:01.280] Yeah. [37:01.660 --> 37:03.020] I know what labels would say. [37:04.800 --> 37:09.260] To be clear, my question is not about whether or not the labels or the people who make money give a fuck about this. [37:09.380 --> 37:09.680] Yeah, yeah. [37:09.680 --> 37:15.400] It's more about whether or not the artists themselves would be tickled by it or if they would be averse to it. [37:15.700 --> 37:17.120] Yeah, no, it's a good question. [37:17.340 --> 37:19.520] I bet that there would be all kinds of reactions. [37:19.900 --> 37:23.220] I think one thing that helps this project is that it's like... [37:24.060 --> 37:25.800] It's not trying to make money off of it. [37:25.860 --> 37:27.560] It's just public shows. [37:31.060 --> 37:34.920] And I think a label would probably say that they don't care about that. [37:37.340 --> 37:38.360] I'd love to... [37:38.360 --> 37:40.660] If anyone knows any of these artists... [37:42.060 --> 37:43.500] I'd love to talk to those artists. [37:45.120 --> 37:45.480] Yes? [37:46.400 --> 37:46.980] Oh, yes. [37:47.140 --> 37:47.760] Okay, go for it. [37:48.220 --> 37:52.880] I just wanted to let you know that I think there's like 18 requests under the demo show ID. [37:52.880 --> 37:53.320] Oh, no. [37:53.760 --> 37:55.200] Oh, it was under demo show. [37:55.380 --> 37:56.640] Yeah, so I just wanted to let you know that. [37:56.640 --> 37:57.680] That's a great... [37:57.680 --> 38:01.320] And some people still managed to figure out how to request it over here. [38:01.480 --> 38:02.040] And I was... [38:02.040 --> 38:04.440] I just wanted to apologize for missing your demo. [38:04.680 --> 38:05.680] Could you play another song? [38:05.820 --> 38:07.120] Oh, we're gonna play some songs. [38:07.200 --> 38:07.500] Oh, good. [38:07.640 --> 38:07.860] Thank you. [38:08.520 --> 38:09.560] Alright, let's do it. [38:11.420 --> 38:12.300] Oh, my God. [38:12.480 --> 38:17.700] Okay, so the lesson here is one of those QR codes turned out to link to the demo show. [38:18.020 --> 38:19.340] And it's my fault. [38:19.720 --> 38:19.920] Okay. [38:21.240 --> 38:23.160] We won't get to all of these songs. [38:23.160 --> 38:25.860] So I'm gonna go through and... [38:25.860 --> 38:26.640] Hmm. [38:26.700 --> 38:29.260] I'm gonna have to exercise some executive judgment here. [38:29.860 --> 38:31.620] This first one is great, actually. [38:31.880 --> 38:34.180] We can go in order of requests by default. [38:34.740 --> 38:39.780] All the small things by blink-182 with replacements from the Big Lebowski script. [38:39.780 --> 38:41.080] Oh, wait. [38:41.320 --> 38:41.800] I... [38:41.800 --> 38:44.760] I will say, the form 1040 is worth seeing. [38:45.000 --> 38:53.340] So, would the person who requested all the small things in Big Lebowski be okay with switching to all the small things with form 1040? [38:56.960 --> 38:57.620] That's true. [38:57.760 --> 38:58.520] They might be in the stream. [38:59.440 --> 39:00.940] If I get a response, I can... [39:00.940 --> 39:01.380] Okay. [39:01.600 --> 39:01.780] I... [39:01.780 --> 39:04.180] I'm gonna just use... [39:04.180 --> 39:06.180] I'm gonna abuse my power and do that. [39:06.180 --> 39:13.620] Um, blink-182, all the small things with replacement lyrics from form 1040, 2024. [39:14.180 --> 39:16.000] Uh, let's do it! [39:16.580 --> 39:16.640] Woo! [39:17.340 --> 39:17.660] Excellent. [39:17.960 --> 39:18.220] I want to go. [39:40.740 --> 39:42.020] Frank's name. [39:42.520 --> 39:42.980] One. [39:56.920 --> 39:58.400] Check only. [40:02.090 --> 40:03.270] I'm at you all. [40:03.690 --> 40:04.010] 27. [40:05.090 --> 40:05.970] I'm at you. [40:43.320 --> 40:44.960] Adline 16 and 17. [40:49.430 --> 40:50.390] It's precious. [40:50.970 --> 40:51.470] 35. [41:08.470 --> 41:09.630] All right. [42:05.900 --> 42:06.820] Oh, that was great. [42:06.940 --> 42:07.200] That was great. [42:07.340 --> 42:07.600] That was great. [42:07.880 --> 42:08.000] That was great. [42:08.120 --> 42:09.500] That was great. [42:09.600 --> 42:10.320] Oh, yeah. [42:10.480 --> 42:12.140] They're all better than the original. [42:13.440 --> 42:14.120] Oh, wait. [42:14.200 --> 42:15.060] We don't do it again. [42:16.400 --> 42:17.140] Not again. [42:17.380 --> 42:17.600] Okay. [42:20.920 --> 42:21.360] Um... [42:21.360 --> 42:22.600] Well, we have someone... [42:23.140 --> 42:23.920] Six more minutes. [42:24.320 --> 42:29.220] So, we have time for probably either one long song or two small songs. [42:30.660 --> 42:31.460] Friday, I'm in Love. [42:31.460 --> 42:31.640] Friday, I'm in Love. [42:31.780 --> 42:33.740] I don't think we have that in the catalog. [42:33.740 --> 42:35.480] We could add it to the catalog. [42:35.780 --> 42:35.920] We could add it to the catalog. [42:35.920 --> 42:39.440] If you have catalog requests, see me after the show for both song and text. [42:40.820 --> 42:41.260] Um... [42:41.260 --> 42:43.800] Oh, we have A Cruel Angel's Thesis requested for everyone. [42:43.960 --> 42:45.120] Does anyone know this song? [42:45.120 --> 42:48.080] This is the Neon Genesis Evangelion theme song. [42:49.380 --> 42:49.740] Uh... [42:49.740 --> 42:49.880] All right. [42:50.000 --> 42:50.520] Let's try that one. [42:50.600 --> 42:52.140] That's short and we'll leave this time for one other. [42:52.320 --> 42:54.280] With the Mean Girls script as replacements. [42:55.240 --> 42:55.600] Uh... [42:55.700 --> 42:56.440] So, this is... [42:57.220 --> 42:58.960] A Cruel Angel's Thesis. [42:59.100 --> 43:01.500] The Neon Genesis Evangelion theme song. [43:01.680 --> 43:03.860] With replacements from Mean Girls script. [43:04.940 --> 43:06.280] If you know it, sing along. [44:39.060 --> 44:39.940] All right, okay. [44:40.160 --> 44:40.640] One more. [44:40.820 --> 44:41.920] And we'll choose this short one. [44:42.720 --> 44:44.160] One more short one. [44:44.160 --> 44:46.020] There's a lot of long ones. [44:46.440 --> 44:46.680] Okay. [44:47.640 --> 44:49.040] Mr. Blue Sky is too long. [44:49.200 --> 44:50.020] Island's in this tree. [44:50.440 --> 44:51.780] Did you see the Rainbow Connection? [44:51.840 --> 44:53.480] Rainbow Connection is a good closer. [44:53.700 --> 44:54.240] Let's do that. [44:54.300 --> 44:54.780] Where's that? [44:54.880 --> 44:55.560] Rainbow Connection. [44:55.580 --> 44:56.080] Scroll up. [44:56.180 --> 44:56.380] Oh. [44:56.540 --> 44:58.100] Oh, yeah. [44:58.300 --> 44:58.440] Okay. [44:58.500 --> 44:59.840] We have two Rainbow Connections. [44:59.980 --> 45:01.780] One is Arrow with Trip Descriptions. [45:02.100 --> 45:04.760] And the other one is, uh, OK Cupid Profiles. [45:04.900 --> 45:05.980] Let's do the OK Cupid. [45:06.500 --> 45:07.880] That's a good, feel good one. [45:08.180 --> 45:08.380] Woo! [45:08.480 --> 45:10.760] Arrow with Trips, we can leave for another day. [45:10.760 --> 45:12.420] Uh, thank you so much. [45:12.620 --> 45:13.640] You've been a great crowd. [45:14.060 --> 45:14.880] Uh, and yeah. [45:15.180 --> 45:18.220] Talk to me if you have suggestions for any part of this. [45:18.640 --> 45:20.740] Specifically, the text data sets. [45:20.860 --> 45:25.220] That's where I'm really trying to get new additions to the catalog these days. [45:25.540 --> 45:25.760] Okay. [45:26.620 --> 45:27.600] Kermit the Frog. [45:27.600 --> 45:28.860] Tick the Frog. [45:28.860 --> 45:28.960] Take this one. [45:31.280 --> 45:34.300] Cupid the Frog. [45:42.760 --> 45:46.120] Cupid the Frog. [45:46.300 --> 45:47.680] Beets. [45:47.940 --> 45:49.000] Henry Burks. [45:49.420 --> 45:52.600] Especially the events. [45:52.600 --> 45:53.520] I love being outside. [45:55.580 --> 46:06.160] Thinking decisions, optical illusions, I guess I love being outside. [46:10.660 --> 46:18.380] I had a chance to keep up with it, while she does the same to me. [46:20.260 --> 46:31.280] Just enjoying it, and with no protection, especially the ocean and sea. [46:36.720 --> 46:46.570] Attempts to find a niche, or playing drums or guitar. [46:48.700 --> 46:52.160] Hi everyone, I'm Matt. [46:52.440 --> 46:54.960] If you can't believe it. [46:56.360 --> 46:57.620] Violet [47:01.970 --> 47:04.530] or scuba diving. [47:05.110 --> 47:07.410] You guys are amazing. [47:07.410 --> 47:12.210] And I'm shining by being me. [47:14.010 --> 47:20.070] Twice if I like it, but a slight direction. [47:20.670 --> 47:24.910] Age is just a number to me. [47:24.910 --> 47:35.690] Then I come out of my shell, though the dancing is far less tragic. [47:37.930 --> 47:41.110] And playful I am deep. [47:41.370 --> 47:43.730] If given to choices. [47:44.490 --> 47:48.810] I'm Nick, but Nick, you exclaim. [47:50.490 --> 47:53.870] Straight out of the ground. [47:54.170 --> 47:56.470] Vintage travel trailers. [47:57.130 --> 48:01.370] So if you want to catch a game. [48:03.410 --> 48:06.510] Closer friends than many. [48:06.910 --> 48:08.970] I can't believe it. [48:09.690 --> 48:13.550] Are as tall as they claim to be. [48:15.590 --> 48:18.550] But when you get in. [48:19.170 --> 48:21.390] And with no protection. [48:22.450 --> 48:26.290] And some fun and garbage TV. [48:27.490 --> 48:31.110] For flavor it's good for some. [48:31.610 --> 48:35.550] I hate java anyway. [48:36.630 --> 48:36.890] Oh! [48:43.660 --> 48:44.400] All right. [48:44.680 --> 48:45.820] Thank you very much. [48:48.440 --> 48:49.560] Yeah, we do this. [48:50.320 --> 48:54.220] Go to robotkaraoke.live and you can find the whole schedule. [48:54.500 --> 48:55.900] Wonderville is the main place we do it. [48:56.380 --> 48:57.280] It's in Brooklyn. [48:58.180 --> 49:01.580] Yeah, it's a Kosciuszko stop on the J. [49:02.280 --> 49:04.020] Yeah, go... [49:05.400 --> 49:06.800] Kosciuszko... I don't know. [49:07.000 --> 49:10.320] The train people on the mics on the PAs say it different ways. [49:10.740 --> 49:11.540] Thank you so much. [49:12.380 --> 49:13.780] Let's hear it again for Jamie! [49:13.780 --> 49:13.900] Thank you.