(having some firefox issues, should work in chrome. will fix firefox asap) posted by dmd at 11:56 AM on February 9
You might want to make that word list a little more readable though if you want people to be able to check for whether the words they want to use are actually in the list. And/or do some slick ajaxy real-time word-completion-list dealiebobber so it's obvious when they're writing a rogue word.
For context for like everyone else, I recorded the first thousand or so words on the mefi word frequency tables back in like 2011 because dmd asked for it on a whim. Apparently he finally did something with that. posted by cortex(staff) at 11:57 AM on February 9
The real-time word-completion-list dealiebobber was originally the idea, which is why I stole the code from up goer five, but I couldn't get the js to work.
Better JS minds than mine (which is basically everyone) are more than welcome to send me patches. posted by dmd at 12:00 PM on February 9
Finally, the fool-proof way to read poetry onstage! posted by The Whelk at 12:08 PM on February 9 [1 favorite]
> For context for like everyone else, I recorded the first thousand or so words on the mefi word frequency tables back in like 2011 because dmd asked for it on a whim. Apparently he finally did something with that.
But what did he do? I don't understand. posted by sweetkid at 12:11 PM on February 9
sweetkid: Type stuff in the box. Press Speak. It reads it in cortex's voice, assuming you used one of the top 1000 words used on metafilter. If it doesn't have the word, it just skips that word.
If anyone can tell me why it doesn't work (sometimes) in Firefox, I'd be grateful. posted by dmd at 12:13 PM on February 9
He cut up all the individual words in the word list I recited, indexed them against the word list, and created a little javascript toy that will speak in a stilted, uninflected, vocabulary-poor simulation of me for arbitrary text entered. posted by cortex(staff) at 12:13 PM on February 9
I like the pauses between words when played on my iPad. It gives tension and deliberation to the sentence being spoken. posted by birdherder at 12:14 PM on February 9
Between the audio from the Metafilter podcasts and the transcripts, one might be able to expand the word list. posted by RichardP at 12:22 PM on February 9
The real challenge is, can you have cortex explain your job using only the ten hundred most common words? posted by LobsterMitten(staff) at 12:28 PM on February 9 [1 favorite]
Leave. The. money. Near. The . Fountain. I. Will. Be. Watching. posted by The Whelk at 12:36 PM on February 9 [1 favorite]
Between the audio from the Metafilter podcasts and the transcripts, one might be able to expand the word list.
Really, if someone wanted to do all the extra work (the bulk of which I think would be in just slicing up the damn word list recording into word-sized bits and indexing them), the simplest thing would be for me to just read more words. It's not terribly exciting stuff but it's not very time consuming either.
Other random thoughts, for potential extenders:
- Starting the playback of sample n a little bit before the end of playback of sample n-1 would pep up playback speed a little and get rid of some of the gaps.
- Having a link generated for a given phrase to make it easy to link would be fun. Right now I guess you'd just have to hack the url manually.
- A failover mode for unidentified words where it plays a short white noise sample or a bleep could help preserve the illusion of flexibility even when it doesn't know a word.
- Parsing some basic punctuation for pauses and maybe rough inflection could be fun: turn commas into short pauses, periods/exclamation points/question marks/colons/semi-colons into longer pauses, and use a pitch shift to turn terminal words down a little and interrogatives up.
Obviously none of that will change it being a fundamentally silly approach to speech synthesis, but hey. posted by cortex(staff) at 12:37 PM on February 9
I just played the Fuck mp3, well said dear cortex posted by wheelieman at 12:55 PM on February 9
I was all disappointed that it wan't working for me, and then figured, ok, maybe lol isn't a commonly used word. Let's take lol out. So I did. But it still didn't work. Typed something else and it did work. So now I'm all disappointed about a whole new thing:
cortex, how come butts isn't one of your most commonly used words? I feel like you've really dropped the ball here. posted by phunniemee at 1:00 PM on February 9
Hey, this was based on aggregate word frequency across the site. If butts is not common enough in that context, the blame falls with everyone but me. posted by cortex(staff) at 1:01 PM on February 9 [7 favorites]
"Dump the motherfucker already" doesn't work either. :( posted by JanetLand at 1:01 PM on February 9 [2 favorites]
Aw, man. Both "plate" and "beans" are missing from the cortex codex. posted by subbes at 2:35 PM on February 9
metafilter is a website that has many links to lots of other ones often with similar or related links put together with each other and where various people post comments to talk about the links
I cannot believe "trigger" and "warning" weren't in there. There goes my plan to record that to be my text message alert when my parents text me. posted by kimberussell at 2:41 PM on February 9 [5 favorites]
Bug report: typing "eight" causes it to say "eighty". posted by Flunkie at 2:47 PM on February 9 [1 favorite]
Oh yes, just in time for the MeFi Swap mix CDs. posted by maryr at 2:52 PM on February 9
How is it possible that "roguelike" is not on this list. posted by ocherdraco at 4:49 PM on February 9
Cortex! Upgrade your version of Dragon Naturally Speaking! It's a lot better now than it was in 1997. posted by double block and bleed at 5:00 PM on February 9
This was just played at the board game meetup I am hosting.
Fortunately for you, the meetup is at my place, so there's not a bar, just beer that's already been paid for. posted by ocherdraco at 5:01 PM on February 9 [1 favorite]
Added a share button. (Thanks tmkf!) posted by dmd at 5:12 PM on February 9 [1 favorite]
Following his existential crisis, Cortex sticks it to the Matt I mean to the Man.
'Death' doesn't sound quite right? Or does it? posted by Wordshore at 5:14 PM on February 9 [1 favorite]
John Madden John Madden Matt Haughey Matt Haughey uiuiuiuiuiuiuiu uiuiuiui John Madden Matt Haughey John Madden Matt Haughey posted by boo_radley at 8:16 PM on February 9 [1 favorite]
cortex: "For context for like everyone else, I recorded the first thousand or so words on the mefi word frequency tables back in like 2011 because dmd asked for it on a whim. "
Ah. I figured this was using the podcast transcripts and some annotations to train up something. posted by pwnguin at 9:50 PM on February 9
That code was way too dense! I ran the code through a beautifier and re-introduced comments. It's in my fork on github. posted by Pronoiac at 12:36 PM on February 10
There's an odd bass tone in the background of the word "this" (and presumably the words preceding/following it in the recording?), any idea what that is?
Also making the audio slices at zero-crossings would seriously improve the flow of the speech. posted by TwoWordReview at 12:22 PM on February 11
TwoWord: If you have the expertise, you're more than welcome to submit a pull request! posted by dmd at 6:15 AM on February 13
I did some work with the audio - trimmed off some heading and trailing silence from the encoder, switching from stereo to mono, and a slight (1/100 second) fade-in and fade-out, which might satisfy TwoWordReview. But both lame and iTunes were adding some padding, which is frustrating. Any thoughts, or any other requested audio tweaks from sox? posted by Pronoiac at 1:12 PM on February 13
Dunno. Put up a fork and try it out? posted by dmd at 7:24 PM on February 13
posted by vrakatar at 11:49 AM on February 9