samples

Strudel's local sample folder (upstream: copy samples here, run npx @strudel/sampler, use samples('local:')).

In jevstrudel it is the sample pack the songs use, served with the site as jev-samples/strudel.json: the static build bundles it, and the local server serves this folder live, so a new sound plays before a rebuild. Each folder is one sound and its files are that sound's variants (s("jeff:0"), …). The per-sound folders carry no README/CLAUDE.md of their own: they are mechanical (folder = sound), and the table below documents all of them.

Sound	What it is
`jeff`	"My name is Jeff" from 22 Jump Street, myinstants `22-jump-street-my-name-is-jeff`, https://www.myinstants.com/media/sounds/22-jump-street-my-name-is-jeff.mp3 (retrieved 2026-09-25; the earlier cut's source, matched by cross-correlation at 0.294 s): 0.294 s to the end of the clip (1.12 s), so the words end at 0.76 s and the room tone decays after them, raised 6.5 dB, 60 ms fade out; −18.5 LUFS, −3.6 dBFS peak. Whisper `small.en` and `medium.en` both read "My name is Jeff." It was a 1 s cut ending mid-room-tone
`mnjev`	"My name is Jev. J, E, V." — Windows SAPI voice (Microsoft David), rate 4, typed as "My name is Jevv. J, E, V." (SAPI's "Jev" is heard as "Jeff"); 2.34 s, 170 ms tail, 20 ms fade, peak −1.9 dBFS (−23.4 LUFS). Whisper `small.en` and `medium.en` both read "My name is Jev. J-E-V." It replaced a cut of "My name is Jev" that both heard as "Jeff" and that stopped mid-/v/
`tsjev`	"Type safe Jev" — same voice, rate 0, typed as "Typesafe Jevv."; 1.11 s, 180 ms tail, peak −2.0 dBFS (−20.3 LUFS). Alone, whisper hears "Type save Jev" (`small.en`, raw render) or "Type safe JEV" (`medium.en`, after 300 ms of silence), but not both on one file: `small.en` reads the cut "type savegev", `medium.en` "Type savejeb". Laid out as build 2 plays it, `medium.en` reads "Type safe Jev"
`ts`	"Type safe" — same voice, rate 0; 0.72 s, 180 ms tail, peak −1.9 dBFS. Whisper `small.en` "type save", `medium.en` "Type safe."
`jev`	"Jev" — same voice, rate 0, a fresh render typed as "Jevv." (not a slice of `mnjev`); 0.53 s, 170 ms tail, peak −1.8 dBFS. Whisper `small.en` reads "Jev."; `medium.en` reads "Jeb?", as it does every spelling tried (Jev, Jevv, Jevvv, Jehv, Jevve, Jev!, other voices)
`rightjump`	"You paid by the billion to press right jump" — SAPI, vocoded on Am (a2 a3 c4 e4); levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1 (was −16.8 LUFS, −1.3 dBTP)
`twotwoseven`	"Two hundred twenty seven milliseconds a move" — SAPI, vocoded on Am; levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1 (was −15.7 LUFS, −1.3 dBTP)
`tictactoe`	"Tic tac toe was solved before you were born" — SAPI, vocoded on F (f2 f3 a3 c4); levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1 (was −17.3 LUFS, −1.3 dBTP)
`joysticks`	"Jev is for judgment, not for joysticks" — re-rendered 2026-09-25 from SSML (`SpeakSsml`, Microsoft David) with Jev as the IPA phoneme `dʒɛv`, vocoded on F (f2 f3 a3 c4), then the dry voice mixed back over its first 0.62 s ("Jev is") with the vocoder ducked 75% under it. Vocoded alone, whisper small.en and medium.en both heard "Jeb" (the old render: "Jeff"); now medium.en reads "Jev is for judgment, not for joysticks", while small.en still hears "Jeff". levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1
`pidloop`	"A PID loop takes the racing line" — re-rendered 2026-09-25 from SSML as "a", a 250 ms `break`, then "PID loop takes the racing line.", vocoded on C (c3 g3 c4 e4), so the article cannot fuse with the letters (the old render read as "APID"). Whisper small.en and medium.en both read "PID loop takes the racing line", hearing the "a" as "Okay". levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1
`sudo`	"Route the ticket. Gate the sudo." — SAPI, vocoded on C; levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1 (was −17.4 LUFS, −1.4 dBTP)
`minimax`	"Minimax was right there" — SAPI, vocoded on G (g2 g3 b3 d4); levelled 2026-09-25 to −16.4 LUFS with peaks at −2.4 dBTP or under (silence trimmed, `volume`, then `alimiter`), like every MINIMAX WAS RIGHT THERE line, so they play alike at gain 1 (was −18.2 LUFS, −1.3 dBTP, the quietest line)
`blueshell`	Mario Kart Wii spiny (blue) shell warning and hit, from YouTube `lw91f4Km6jE`; trimmed from 0.626 s so the hit lands 2 bars in at 170 bpm; raised to −15.5 LUFS (was −19.6)
`ohno`	"Oh no no no" (the laugh version), myinstants `oh-no-no-no-no-laugh`, first 2.15 s
`newticket`	"New ticket in" — SAPI, vocoded on Cm9 (c3 eb3 g3 bb3 d4)
`nohuman`	"No human in the loop" — SAPI, vocoded on Cm9
`gotjevcm`	"You've got Jev" — SAPI, vocoded on Cm9; re-exported 2026-09-25 with `ffmpeg -af loudnorm=I=-16:TP=-1.5:LRA=7` (64 kb/s mono), since it reached +0.3 dBTP in the mix: now −16.9 LUFS, −2.0 dBTP
`billingbug`	"Billing, bug, or refund" — SAPI, vocoded on Fm9 (f2 ab3 c4 eb4 g4)
`routedblink`	"Routed in a blink" — SAPI, vocoded on Fm9
`gotjevfm`	"You've got Jev" — SAPI, vocoded on Fm9; re-exported 2026-09-25 with `ffmpeg -af loudnorm=I=-16:TP=-1.5:LRA=7` (64 kb/s mono), since it reached +0.3 dBTP in the mix: now −16.4 LUFS, −2.0 dBTP
`choicebilling`	"Choice: billing" — SAPI, vocoded on Bbmaj7 (bb2 d3 f3 a3)
`nobodyread`	"Nobody had to read it" — SAPI, vocoded on Bbmaj7; −18.4 LUFS, about 1–2 dB under the other verse lines, so YOU'VE GOT JEV plays it at gain 1.35 against their 1.1
`ninetythree`	"Point nine three" — SAPI, vocoded on Abmaj7 (ab2 c3 eb3 g3)
`gotjevab`	"You've got Jev" — SAPI, vocoded on Abmaj7; re-exported 2026-09-25 with `ffmpeg -af loudnorm=I=-16:TP=-1.5:LRA=7` (64 kb/s mono), since it reached +0.3 dBTP in the mix: now −17.1 LUFS, −2.0 dBTP
`mail`	AOL "You've got mail", myinstants `aol-you-got-mail`, 0.3–1.95 s (the words); loudness-normalised to −16 LUFS (`ffmpeg -af loudnorm=I=-16:TP=-1.5:LRA=7`) to match the `gotjev*` lines: the source was −7 LUFS and clipped at +1.5 dBFS, about 9 dB over the Jev replies it alternates with
`modem`	A dial-up modem dialing in, "Dial up modem noises" by William Termini, public domain, https://commons.wikimedia.org/wiki/File:Dial_up_modem_noises.ogg (retrieved 2026-09-25): 0.95–4.1 s (dial tone 350+440 Hz, then DTMF digits) joined to 10.0–14.3 s (the 2100 Hz answer tone into the handshake), 7.45 s, raised 6.5 dB to −17.5 LUFS (−2.4 dBTP) to match the clip it replaced. Bands measured per 0.25 s window with an FFT. It replaced myinstants `aoldialup1` 0–14.3 s, which opens on the answer tone with no dial tone or dialing
`screech`	AOL dial-up, myinstants `aoldialup1`, 14.3–18.7 s: the handshake screech
`goodbye`	AOL's "Goodbye", Elwood Edwards (the voice of AOL's "You've got mail"; https://en.wikipedia.org/wiki/Elwood_Edwards), `Goodbye.wav` from the Internet Archive's "AOL Instant Messenger sounds", https://archive.org/details/im_20191103 (retrieved 2026-09-25): the whole 0.51 s clip, which whisper.cpp (`ggml-base.en`) reads as "Goodbye" from its start; lowered 5.2 dB to −16.2 LUFS (−5.6 dBTP) to match `mail`, resampled to 22.05 kHz. It replaced an unidentified voice from myinstants `goodbye_ZyZcmQs`
`kartfall`	Mario Kart "fall off cliff", myinstants `fall-off-cliff`, 0.2–1.9 s
`sothinking`	"Thinking, fast and slow." — SAPI (Microsoft David), plain, not vocoded; raised to −17.0 LUFS (gain + limiter at −1.5 dBTP): it was −21.2, 4 dB under the hook
`sohook`	"Noul, choice, score." — re-rendered 2026-09-25 so "Noul" reads as one word: SAPI (Microsoft David), rate −4, typed as "Noul, choice, score." (the product's own spelling), vocoded on Em (e2 e3 g3 b3 e4) with `--dry 3`, the pauses between the words cut to 0.24 s; 2.88 s, −17.9 LUFS, −1.9 dBFS peak (gain + limiter). Dry, whisper small.en reads "Newel. Choice. Score." and medium.en "Null, choice, score."; in SYSTEM ONE's mix (recorded headless, drops 1 and 2 at gain 2.2) small.en reads "Newel Choice Score" / "Newel. Choice. Score." and medium.en "Newable. Choice. Score." / "No choice. Score.": one word every time. The old render (typed "Nool", rate 2, `--dry 1.5`) read "No all" / "Know all", dry and in the mix, as did "Nool" at rate −4
`socook`	"Let him cook!", myinstants `let-him-cook-now`, 0.05–2.25 s; lowered to −18.3 LUFS: it was −12.4 and clipped (+1.7 dBTP), 6–8 dB over the other lines as heard at gain 1.3; the source's clipping remains
`socooknow`	"Let him cook now", same clip, 2.22–3.35 s; lowered to −18.3 LUFS: it was −11.6 and clipped (+1.1 dBTP); the source's clipping remains
`sowinchime`	Windows' own "chimes" system sound, `C:\Windows\Media\chimes.wav` from this machine's Windows 11 (sha256 `bbee519c…7e6d`, retrieved 2026-09-25): the whole 1.23 s, a quick E6/A6/D6 arpeggio settling on A5, summed to mono and raised 23.7 dB to −18.0 LUFS (−2.4 dBFS peak; the source is −36.8 LUFS). Its root measures A5 +8 cents (MIDI 81.08, FFT), so SYSTEM ONE plays it with `note(x).transpose(-45).speed(.995)` to sound on `x`
`hljevplays`	"Jev plays Link to the Past" — SAPI (Microsoft David, rate −4, spelled "Jehv, plays…"), vocoded on Gm (g2 g3 bb3 d4) with `--dry 1.5 --seed 1`, the pause after "Jev" cut to 0.1 s and sped up ×1.1; 2.41 s, −18.4 LUFS. whisper small.en and medium.en read "Jev plays Link to the Past" (base.en hears "Jeb")
`hlnative`	"Native now, WASM some day" — SAPI (rate −1), vocoded on Gm with `--dry 0.3`, the pause after "now" shortened by 0.3 s; 2.49 s, −16.9 LUFS. whisper small.en and medium.en read it back word for word
`hlcore`	"The core's a library" — SAPI, vocoded on Eb (eb2 eb3 g3 bb3)
`hlsame`	"Same program, both" — SAPI, vocoded on Eb
`hlframe`	"Three questions a minute" — SAPI, vocoded on F (f2 f3 a3 c4); 1.16 s, −17.3 LUFS. whisper small.en, base.en and medium.en read it back
`hlheyvoc`	"Hey, listen" — SAPI, vocoded on F
`hlpress`	"The bot walks, Jev picks" — SAPI (rate −4, "The bot walks. Jehv, picks."), vocoded on D (d2 d3 f#3 a3) with `--dry 1.5 --seed 1`, pauses shortened and sped up ×1.06 to fit two bars; 2.80 s, −17.5 LUFS. whisper small.en and medium.en read "The bot walks, Jev, pix"
`hljevplaysd`	"Jev plays Link to the Past" — the speech of `hljevplays`, vocoded on D with `--dry 1.5 --seed 1`, cut and sped up the same way; 2.40 s, −18.4 LUFS. whisper small.en and medium.en read Jev
`hlheylisten`	Navi's "Hey! Listen!", myinstants `hey_listen`, from 0.1 s; leveled to −17.5 LUFS (was −24.4, 7 dB under the verse lines)
`hlhey`	Navi's "Hey!" alone, same clip, 0.1–0.6 s with a short fade; leveled to −17.5 LUFS (was −22.1)
`hldanger`	"It's dangerous to go alone! Take this." — a fan reading of the 1986 game's on-screen text, YouTube `M5JJCBQlqa0` (Eccentric Sounds, "Copyright Free Sound Effect", 2025-03-22), 1.02–4.20 s with 10 ms / 50 ms fades, +1.5 dB to −17.3 LUFS. Re-sourced 2026-09-25: the best-known clip, myinstants `its-dangerous-to-go-alone-take-this`, swallows "alone" (whisper small.en and medium.en both read "to go, but don't take this") and the old cut at 2.45 s ended inside "this"; this one reads back word for word on both
`tcinput`	"Input tokens, forty-two per billion" — SAPI, vocoded on Am7 (a2 e3 g3 c4); raised 3.2 dB to −17.4 LUFS
`tcroute`	"Route the ticket" — SAPI, vocoded on Am7
`tcfreeam`	"Output tokens: free" — SAPI, vocoded on Am7; raised 2.6 dB to −17.3 LUFS
`tcfree`	"Output tokens: free" — SAPI, vocoded on D9 (d3 f#3 a3 c4 e4); raised 4.0 dB to −17.4 LUFS
`tcflag`	"Flag the jailbreak" — SAPI, vocoded on D9
`tcseventy`	"Seventy to five hundred milliseconds" — SAPI, vocoded on Gmaj7 (g2 d3 f#3 b3); raised 1.7 dB to −17.1 LUFS
`tcgate`	"Gate the sudo command" — SAPI (Microsoft David) at rate -2, spelled "Gate, the sue-doo command" so "sudo" is not heard as "pseudo" and the comma gives "Gate" room, vocoded on Gmaj7 (g2 d3 f#3 b3), silence trimmed, −16.1 LUFS; re-rendered 2026-09-25. whisper.cpp (`ggml-base.en`) reads "Gate. The sudo command." dry and over the four-on-the-floor kick at the vocal's level and 4.4 dB above it; the render it replaced read "Gate the pseudo command" dry and "Hate the pseudo command" over the kick
`tcfreeg`	"Output tokens: free" — SAPI, vocoded on Gmaj7; raised 2.4 dB to −17.2 LUFS
`tcmeter`	"Too cheap to meter" — SAPI, vocoded on Cmaj7 (c3 e3 g3 b3)
`tcrealestate`	"It's free real estate" (Tim and Eric), myinstants `its-free-real-estate`, 0–1.2 s; raised 7.5 dB under a limiter to −17.2 LUFS (was −24.0)
`lbquote`	"We have lightning in a bottle, and yet it is not useful." — SAPI, plain (not vocoded), 3.06 s; raised to −17.1 LUFS, limited to −1.5 dBTP (was −19.7)
`lbremove`	"Remove the human from the loop" — SAPI, vocoded on Em (e2 b2 e3 g3 b3); 1.34 s, −15.5 LUFS, −1.3 dBTP. whisper small.en and medium.en read "Remove the human from the loop."
`lbcodeem`	"The code decides what code can decide" — SAPI, vocoded on Em; 2.04 s, −16.5 LUFS, −1.3 dBTP. whisper small.en and medium.en read it word for word
`lbroutine`	"Routine calls never leave the machine" — SAPI, vocoded on C (c3 e3 g3 c4); 1.85 s, −16.6 LUFS, −1.3 dBTP. whisper small.en and medium.en read it word for word
`lbcodec`	"The code decides what code can decide" — SAPI, vocoded on C; 2.04 s, −18.0 LUFS, −1.2 dBTP. whisper small.en and medium.en read it word for word
`lbresidue`	"Only the residue goes to Jev" — re-rendered 2026-09-25: SAPI (Microsoft David, rate −4, typed "Only the residue goes to Jevv, "), vocoded on G (g2 d3 g3 b3) with `--dry 1.5 --seed 1`, silence trimmed, +5 dB into `alimiter` at 0.63, 20 ms fade; 2.83 s, −16.8 LUFS, −3.7 dBTP. whisper small.en and medium.en read "Only the residue goes to Jev." (the old render: both heard "Jeff"; typed "Jev" or "Jehv" at rate −4, small.en still does)
`lbrestg`	"Jev takes the rest" — re-rendered 2026-09-25: SAPI (rate −4, typed "Jevv. Takes the rest."), vocoded on G with `--dry 1.5 --seed 1`, the pause after the name cut from 1.37 s to 0.25 s, +5 dB into `alimiter` at 0.63, 20 ms fade; 2.45 s, −17.5 LUFS, −3.8 dBTP. whisper small.en reads "Jev, takes the rest.", medium.en "Jev takes the rest." (the old render: "Jeff"/"Jeb"; typed "Jevv, takes" with a comma, small.en still reads "Jeb")
`lbfortyfour`	"Forty-four out of forty-four" — SAPI, vocoded on D (d3 f#3 a3 d4); 1.52 s, −17.6 LUFS, −1.4 dBTP. whisper small.en and medium.en read "44 out of 44"
`lbrestd`	"Jev takes the rest" — the speech of `lbrestg`, vocoded on D with `--dry 1.5 --seed 1`, cut and levelled the same way; 2.45 s, −17.9 LUFS, −3.6 dBTP. whisper small.en reads "Jev, takes the rest.", medium.en "Jev takes the rest."
`lbshotgun`	Doom shotgun fire + pump, myinstants `doom-shotgun-2-0`, from 0.03 s; lowered to −15.1 LUFS, −3.0 dBTP (was −5.4 LUFS, clipping at +6.5 dBTP, 11 dB over the vocals)
`fffortyfive`	"Forty-five percent" — SAPI, vocoded on Fmaj7 (f2 a3 c4 e4); levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping (was −17.6 LUFS, −1.3 dBFS peak)
`ffupflat`	"Up, flat, or down" — SAPI (text "Up, flat, or down.", 2.0 s), vocoded on Em7 (e2 g3 b3 d4), line 2 of verse 1, sung a beat after the bar; levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping
`fflayer`	"A disciplined reasoning layer" — SAPI, vocoded on Fmaj7; levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping (was −15.9 LUFS, −1.3 dBFS peak)
`ffcoin`	"Better than a coin" — SAPI, vocoded on Dm7 (d2 f3 a3 c4), line 3 of verse 1 (re-rendered 2026-09-25 when it moved from Em7); levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping
`ffnotmodel`	"Not a trading model" — SAPI, vocoded on Em7 (e2 g3 b3 d4); levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping (was −17.3 LUFS, −1.4 dBFS peak)
`ffindex`	"Worse than an index fund" — SAPI, vocoded on Cmaj7 (c3 e3 g3 b3), line 4 of verse 1 (re-rendered 2026-09-25 when it moved from Dm7); levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping
`fftrust`	"Don't trust the word" — SAPI, vocoded on Dm7; levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping (was −16.9 LUFS, −1.4 dBFS peak)
`ffprob`	"Probability" — SAPI, vocoded on Cmaj7; levelled to −17.3 LUFS with its peaks limited to −4.5 dBFS (`volume`, then `alimiter`), like every FORTY-FIVE PERCENT verse line, so the vocals sit forward without the mix clipping (was −17.5 LUFS, −1.8 dBFS peak)
`ffstonks`	"Stonks", myinstants `stonks_jlBhTbJ`, 0.05–1.0 s; loudness-normalised to −16.7 LUFS (`loudnorm=I=-16.5:TP=-1.5`): the source was −6.5 LUFS and clipped at +3.6 dBFS, about 10 dB over the vocal lines it plays among
`ffhum`	Wolf of Wall Street chest-beat humming, YouTube `XYVXOky-j5g`, 13.0–16.0 s, re-cut 2026-09-25 into one bar at 80 bpm (3.0 s) of eight 8th-note slices, each starting on a thump: the thump at 0.127 s with its hum tuned +32 cents to A2 (110 Hz; slices 0 and 4), the thump at 1.596 s tuned +16 cents to E2 (82.4 Hz; slices 1 and 5), the A grain shifted down a fifth to D2 (slices 2 and 6), and the A grain cut to its 0.14 s thump (slices 3 and 7). Each grain tuned by its own median pitch (YIN) with `asetrate`/`atempo` and faded out within its slice; peak −3 dBFS. The source was about 30 cents flat, slid through G#, drifted off the 8th grid and faded 10 dB over its last 3/8; its thump at 0.613 s reads a semitone sharp and is left out
`fffine`	"This is fine", myinstants `fine`, 3.9–5.67 s (after the clip's crickets), 80 ms fade-in; loudness-normalised to −16.7 LUFS (`loudnorm=I=-16.5:TP=-1.5`): the source was −23.3 LUFS, about 6 dB under the vocal lines, and it is the song's last word
`ssselect`	"Select star from Jev" — SAPI (Microsoft David, rate 0) spelled "Select star from Jehv", vocoded on F#m (f#2 f#3 a3 c#4) with `--dry 1.5 --seed 1`, silence trimmed (1.31 s), +4.6 dB under a limiter at −1.5 dBFS; −17.1 LUFS, −1.7 dBFS peak. Re-rendered 2026-09-25: whisper `small.en` reads "Select star from Jeff" and `medium.en` "Select star from Jev." The render it replaced ("Jev", no dry voice) read "Select star from GIF" on both; high-passing the dry voice above 4 kHz instead still read "JAV" on `medium.en`
`ssorderby`	"Order by probability" — SAPI, vocoded on F#m
`ssvibes`	"Where vibes are true" — SAPI, vocoded on D (d2 d3 f#3 a3)
`ssforty`	"Forty rows a batch" — SAPI, vocoded on D
`ssscan`	"Custom scan" — SAPI, vocoded on A (a2 a3 c#4 e4)
`ssoptions`	"Two hundred fifty-five options" — SAPI, vocoded on A
`sspool`	"Warm pool, cached answers" — SAPI, vocoded on E (e2 e3 g#3 b3)
`sstree`	"Then a label tree" — SAPI, vocoded on E
`ssvibese`	"Where vibes are true" — SAPI, vocoded on E (chorus bar 23); raised to −17.4 LUFS (gain + limiter at −1.5 dBTP): it was −18.8
`sshellothere`	Obi-Wan "Hello there", myinstants `obi-wan-hello-there`, whole clip (0.82 s); raised to −17.6 LUFS (gain + limiter at −1.5 dBTP): it was −25.9, about 9 dB under the verse lines
`sskenobi`	General Grievous "General Kenobi", myinstants `general-kenobi-sw`, whole clip (1.95 s); raised to −17.2 LUFS (gain + limiter at −1.5 dBTP): it was −24.2, about 7 dB under
`ssgate`	The gated snare's impulse response for SELECT * FROM JEV (`.ir('ssgate')`): 250 ms of flat stereo white noise (numpy, seed 1, peak −6 dBFS), cut dead by a 5 ms linear fade at its end, 44.1 kHz 16-bit WAV. Made here, not sampled: a flat tail that stops, the gated-reverb sound, where the built-in room decays exponentially
`ooqfair`	"Fair or unfair?" (typed "Fair, or unfair?") — SAPI (Microsoft David), vocoded on Dm (d2 d3 f3 a3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 1.59 s. Whisper `small.en` and `medium.en` read "Fair or unfair?"
`ooqunfair`	"Unfair or fair?" (typed "Unfair, or fair?") — SAPI (Microsoft David), vocoded on Bb (bb1 bb2 d3 f3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 1.58 s. Whisper `small.en` and `medium.en` read "Unfair, or fair." / "Unfair or fair?"
`ooqapprove`	"Approve or fix now?" (typed "Approve, or fix now?") — SAPI (Microsoft David), vocoded on F (f2 f3 a3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 1.82 s. Whisper `small.en` and `medium.en` read "Approve or fix now." / "Approve, or fix now."
`ooqfix`	"Fix now or approve?" (typed "Fix now, or approve?") — SAPI (Microsoft David), vocoded on C (c3 e3 g3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 1.70 s. Whisper `small.en` and `medium.en` read "Fix now or approve." / "Fix now, or approve."
`ooafairdm`	"Fair." — the chorus answer — SAPI (Microsoft David), vocoded on Dm (d2 d3 f3 a3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.48 s. Whisper `small.en` and `medium.en` read "Fair."
`ooaunfairbb`	"Unfair." — the chorus answer — SAPI (Microsoft David), vocoded on Bb (bb1 bb2 d3 f3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.66 s. Whisper `small.en` and `medium.en` read "Unfair."
`ooaapprovef`	"Approve." — the chorus answer — SAPI (Microsoft David), vocoded on F (f2 f3 a3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.66 s. Whisper `small.en` and `medium.en` read "Approve."
`ooafixc`	"Fix now." — the chorus answer — SAPI (Microsoft David), vocoded on C (c3 e3 g3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.64 s. Whisper `small.en` and `medium.en` read "Fix now." / "Fix Now"
`ooaunfairdm`	"Unfair." — the outro's swapped answer — SAPI (Microsoft David), vocoded on Dm (d2 d3 f3 a3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.68 s. Whisper `small.en` and `medium.en` read "unfair." / "Unfair."
`ooafairbb`	"Fair." — the outro's swapped answer — SAPI (Microsoft David), vocoded on Bb (bb1 bb2 d3 f3), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.46 s. Whisper `small.en` and `medium.en` read "Fair."
`ooafixf`	"Fix now." — the outro's swapped answer, spoken at rate −2 and vocoded with `--dry 0.8` (the rate-0 render read as "Thanks, Neil" on `medium.en`) — SAPI (Microsoft David), vocoded on F (f2 f3 a3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.75 s. Whisper `small.en` and `medium.en` read "Fix now!" / "Fix now."
`ooaapprovec`	"Approve." — the outro's swapped answer — SAPI (Microsoft David), vocoded on C (c3 e3 g3 c4), rendered 2026-09-25; silence trimmed, levelled to −17 LUFS (`volume`, then `alimiter` at −1.5 dBTP); 0.66 s. Whisper `small.en` and `medium.en` read "Approve."
`oothirtydm`	"Thirty-two out of two hundred" — SAPI, vocoded on Dm
`oothirtyf`	"Thirty-two out of two hundred" — SAPI, vocoded on F
`ooshuffled`	"Wait, it's all shuffled?" — plain SAPI voice (Microsoft David), not vocoded; raised to −17.8 LUFS (gain + limiter at −1.5 dBTP): it was −19.4
`ooalways`	"Always has been" — the spoken meme line, cut 8.15–9.4 s from YouTube `QzOxIqanDXs` ("Wait it's all Ohio?"), with a 0.15 s fade before the choir comes in; raised to −17.6 LUFS (gain + limiter at −1.5 dBTP): it was −21.2
`bcborrow`	"Cannot borrow as mutable" (rustc 1.98.1's E0596 span label) — SAPI, vocoded on Em7 (e3 g3 b3 d4); whisper small.en and medium.en both read "cannot borrow as mutable"
`bctypes`	"Mismatched types" (rustc E0308) — SAPI, vocoded on Em7
`bclintem`	"Lint. Before you spend." — SAPI, vocoded on Em7; the full stop keeps "Lint" from running into "before" and being heard as "Link"; whisper small.en reads "Lint, before you spend", medium.en "Lint. Before you spend"
`bclifetime`	"Lifetime may not live long enough" (rustc) — SAPI, vocoded on A7 (a2 e3 g3 c#4)
`bcshared`	"Which is behind a shared reference" (rustc E0507) — SAPI, vocoded on A7
`bclinta`	"Lint. Before you spend." — SAPI, vocoded on A7 with `--noise 0.5 --noise-above 1800 --env-hz 50` (the default setting is heard as "Lindt"); whisper small.en and medium.en both read "Lint, before you spend"
`bcxlive`	"x does not live long enough" (rustc E0597) — SAPI, vocoded on Dmaj7 (d3 f#3 a3 c#4)
`bcfoobar`	"Expected Foo, found Bar" (rustc E0308 label) — SAPI, vocoded on Dmaj7
`bcmoved`	"Use of moved value" (rustc E0382) — SAPI, vocoded on Bm7 (b2 d3 f#3 a3)
`bctwice`	"Cannot borrow as mutable more than once at a time" (rustc 1.98.1's E0499, without the variable name) — SAPI, vocoded on Bm7; whisper small.en and medium.en both read it back word for word
`bcfinished`	"Finished dev profile" (cargo 1.98's Finished `dev` profile line) — SAPI, each word spoken alone, trimmed and joined with 60 ms gaps (spoken as one line, "dev" is heard as "their", "temp" or "dead"), vocoded on A7; 1.41 s; whisper small.en and medium.en both read "finished dev profile"
`bcfinishedd`	The same words as `bcfinished`, vocoded on Dmaj7 (d3 f#3 a3 c#4) for the outro; whisper small.en and medium.en both read "finished dev profile"
`bcwow`	Owen Wilson "Wow", myinstants `owen_wilson_says_wowmp3converter`, 0–1.7 s; raised 6.8 dB with a limiter at −1.5 dBFS (`volume,alimiter`) to −17.7 LUFS (was −23.3, about 6 dB under the chorus lines)
`tfnohalluc`	"No hallucinations" — SAPI, vocoded on C (c3 e3 g3 c4); Whisper `small.en` and `medium.en` both read "No hallucinations."
`tforderby`	"Order by probability" — SAPI, vocoded on C; Whisper `small.en` and `medium.en` both read "Order by probability."
`tftypesafe`	"Jev is type safe. Jev is fine." — SAPI (Microsoft David, rate −4, typed "Jehv, is type safe. Jehv, is fine."), silence trimmed, vocoded on C (c3 e3 g3 c4) with `--dry 1.5 --seed 1`, then sped up ×2.36 (`atempo`) to 2.59 s so it fits two bars at 180 bpm; levelled 2026-09-25 to −17.3 LUFS, peak −2.2 dBFS (`volume`, then `alimiter`). Whisper `small.en` and `medium.en` both read "Jev is type safe. Jev is fine." The earlier render (rate 0, "Jev") read "Jeff is type safe" in both
`tfwrongopt`	"Just the wrong option" — SAPI, vocoded on G (g2 d3 g3 b3); Whisper `small.en` and `medium.en` both read "Just the wrong option."
`tffourofsix`	"Four of six failed" — SAPI, vocoded on G; Whisper `small.en` and `medium.en` both read "Four of six failed."
`tfranked`	"Ranked the files" — SAPI (rate −2) rendered as two phrases, "Ranked." and "the files.", each trimmed and joined with an 80 ms gap so the final -ed is heard, vocoded on Am (a2 e3 a3 c4) with `--dry 1.5 --seed 1`; 1.45 s, −17.4 LUFS, peak −2.2 dBFS. Whisper `small.en` and `medium.en` both read "Ranked the files." The earlier render read "Rank the files" in both
`tfreversed`	"Reversed the options" — SAPI (rate 0) rendered as two phrases, "Reversed" and "the options.", joined with a 50 ms gap so the final -ed is heard, vocoded on Am (a2 e3 a3 c4) with `--dry 1.5 --seed 1`; 1.31 s, −17.2 LUFS, peak −2.4 dBFS. Whisper `small.en` and `medium.en` both read "Reversed the options." It replaced `tfshuffled` ("Shuffled the options", read as "Shuffle the options"): jev-mcp reversed the options, it did not shuffle them
`tfnoise`	"Indistinguishable from noise" — SAPI, vocoded on F (f2 c3 f3 a3); Whisper `small.en` and `medium.en` both read "Indistinguishable from noise."
`tfthirtytwo`	"Thirty-two of two hundred" — SAPI (rate 0, typed "Thirty two of two hundred."), vocoded on F (f2 c3 f3 a3) with `--dry 1.5 --seed 1`; 1.44 s, −17.0 LUFS, peak −2.3 dBFS. Whisper `small.en` and `medium.en` both read "32 of 200." It replaced `tfflipped` ("Thirty two flipped"): jev-mcp flipped 32 of 200 choices
`tftaskfailed`	"Task failed successfully" — SAPI (rate 0), plain (not vocoded): myinstants has no audio clip of this image meme; re-rendered 2026-09-25, 1.57 s ("successfully" ends at 1.56 s), levelled to −17.1 LUFS, peak −2.2 dBFS. Whisper `small.en` and `medium.en` both read "Task failed successfully." The earlier render, raised 3.3 dB under a limiter, read "The task failed successfully" in `small.en`
`tfxperror`	Windows XP Error, myinstants `erro` ("Windows XP Error"), full 0.99 s; measured centred on C (C4 → G3 → C3)
`tfemodamage`	"Emotional damage" — Steven He, "Asian Parent Punishments: Emotional Damage", YouTube `cQpq56FmIN4` via yt-dlp (retrieved 2026-09-25), 13.75–15.55 s (his first "emotional damage"; the source peaks at −2.2 dBFS with no clipped samples), 60 ms fade-out; 1.80 s, levelled to −17.0 LUFS, peak −2.5 dBFS. Whisper `small.en` and `medium.en` both read "Emotional damage." The earlier myinstants cut was clipped and read "Emotion, oh damn!" and "emotional time"; the phrase in "When 'Asian' Is a Difficulty Mode" (`miD_TWmdGIY`) clips too (18 full-scale samples)
`tfohno`	"Oh no no no", the words alone — myinstants `oh-no-no-no-no-laugh`, https://www.myinstants.com/media/sounds/oh-no-no-no-no-laugh.mp3 (retrieved 2026-09-25; the source of `ohno`), 1.29–2.60 s: whisper hears the first 1.29 s as laughter in both models ("giggles", "Laughing"), so the cut starts where the words do; 10 ms fade in, 60 ms out; 1.31 s, levelled to −17.0 LUFS, peak −3.4 dBFS. Whisper `small.en` and `medium.en` both read "No, no, no, no." It fills one bar at 180 bpm
`tfweide`	"Directed by Robert B. Weide" theme, myinstants `directed-by-robert-b-weide-theme-meme`, full 10.2 s; limited to −1.9 dBTP (was +0.7), −20.5 LUFS
`emcurl`	"Curl, pipe, bash" — SAPI (rate 0), vocoded on Fm (f2 c3 f3 ab3), `--noise 0.6 --noise-above 2500`; whisper.cpp `small.en` and `medium.en` both read "Curl, pipe, bash." (2026-09-25)
`emremote`	"Remote code, point nine nine" — SAPI, vocoded on Db (db2 ab2 db3 f3), whisper setting `--noise 1.0 --noise-above 1200`; whisper.cpp `small.en` and `medium.en` both read "Remote code 0.99" (2026-09-25)
`emdenied`	"Denied" — SAPI (Microsoft David), typed "Denied!" at rate −6, vocoded on Ab (ab2 eb3 ab3 c4), `--noise 0.4 --noise-above 2500 --dry 0.6` (seed 0), silence trimmed, loudness-normalised (`loudnorm=I=-16.8:TP=-1.5`) to −17.2 LUFS, peak −2.5 dBFS; 0.77 s; re-rendered 2026-09-25. Whisper.cpp `small.en` and `medium.en` both read "Denied."; the render it replaced (rate −4, `--noise 0.6`) read "Goodnight." in `small.en`
`emnotimp`	"Jev is not the impostor" — SAPI (Microsoft David), typed "Jehv is not the impostor." at rate 0, vocoded on Eb (eb2 bb2 eb3 g3), `--noise 0.6 --noise-above 2500 --dry 1.5` (seed 0), silence trimmed, loudness-normalised (`loudnorm=I=-17.1:TP=-1.5`) to −19.8 LUFS, peak −2.0 dBFS (the song raises it with gain); 1.30 s, under one bar at 174 bpm; re-rendered 2026-09-25. Whisper.cpp `small.en` reads "Jev is not the imposter.", `medium.en` "Jev is not the impostor." It is fragile: seeds 1–3 of the same render, or a limiter to raise it, read "Jeff" in `small.en`; typed "Jevv" (rates −2 to +2, with and without a comma, the name spliced slower) read "Jebb"/"Jeb"/"Jeff" at one bar's length. The render it replaced (`--noise 1.0 --noise-above 1200`) read "Jeff is not the impostor" in both
`emalarm`	Among Us emergency-meeting alarm, myinstants `alarm_emergencymeeting-online-audio-converter`, 0–2.4 s (blast and tail); loudness-normalised to −13.7 LUFS (`loudnorm=I=-13.8:TP=-1.5`), was −11.8
`emrunning`	"Boi, why are you running?" — from the Nollywood film Pretty Liars 1 (2010), spread as a Vine in 2016 (https://knowyourmeme.com/memes/why-are-you-running, retrieved 2026-09-25); myinstants `boi-why-are-you-running`, 0.68 s to the end (2.64 s), so the speech (onset 0.71 s at −40 dB) starts on the downbeat, 10 ms fade in; loudness-normalised (`loudnorm=I=-15.4:TP=-1.5`) to −15.8 LUFS, peak −4.4 dBFS; 1.96 s; re-cut 2026-09-25 (it was 0–2.64 s, with 0.7 s of silence before the line). Whisper.cpp `small.en` and `medium.en` both read "Boy, why are you running?"
`emsus`	The drawn-out "sus", myinstants `ayo-thats-kinda-sus`, 1.72–3.05 s; loudness-normalised to −15.6 LUFS (`loudnorm=I=-15.7:TP=-1.5`), was −11.8. Whisper.cpp `small.en` and `medium.en` both read it "SIZE!"; the whole source clip reads "Ayo, that's kinda size!" (`small.en`) and "…kinda sus!" (`medium.en`). Re-cuts from "kinda" (1.13 s) to the end of the vowel (2.97 s), with the 0.36 s pause shortened and the vowel sped up to fit a bar, read worse ("Nut die!", "Nerd's eyes!"), as did EQ boosts above 3 kHz, so this cut stays (checked 2026-09-25)
`etvloan`	"Your loan has expired." — the store: SAPI Zira, rate −1, band-passed 300–3400 Hz (two 2-pole high- and low-passes) and bit-crushed (`acrusher=bits=8:mode=log:aa=1:samples=2`), silence trimmed, +8.5 dB with a −1.5 dBFS limiter; 1.3 s; −16.2 LUFS
`etvlog`	"Browser log, supplemental. I have navigated to page four thousand one hundred and twelve of the catalog." — SAPI David, rate +1, vocoded on e3 (`--dry 0.6`); −16.5 LUFS
`etvvolumes`	"I have acquired eleven thousand volumes. I am unable to open any of them." — SAPI David, rate +1, vocoded on e3 (`--dry 0.6`); −16.6 LUFS
`etvofcourse`	Data: "Of course." alone (TNG "The Naked Now"), YouTube `9ev1ec0Z0GI`, 87.40–87.96 s (whisper word times and a 10 ms RMS envelope: "of" starts 87.41, "course" ends 87.93, the next word 88.02), 10 ms fade in, 30 ms out, +24.9 dB with a −1.5 dBFS limiter; 0.56 s; −16.1 LUFS
`etvdenied`	"Access denied." — the store: SAPI Zira, rate −1, band-passed and bit-crushed as `etvloan`, +7.5 dB with a −1.5 dBFS limiter; 1.0 s; −15.8 LUFS
`etvchirp`	TNG combadge chirp, TrekCore `https://www.trekcore.com/audio/communicator/tng_chirp_clean.mp3`, whole clip; peak −3.3 dBFS
`etvbeep`	TNG computer beep, TrekCore `https://www.trekcore.com/audio/computer/computerbeep_1.mp3`, whole clip; −18.5 LUFS
`etvclick`	One mouse click, YouTube `PmaQ_qMko7s`, trimmed, +8.6 dB; peak −4.3 dBFS
`etv429`	"Four twenty-nine. Too many requests." — the store: SAPI Zira, rate −1, band-passed and bit-crushed as `etvloan`, +9.7 dB with a −1.5 dBFS limiter; 3.3 s; −16.3 LUFS
`etvyes`	"Yes." — Jev: SAPI Zira, rate +3, silence trimmed, +8.1 dB with a −2 dBFS limiter; 0.36 s; −16.8 LUFS (looped)
`etvshould`	"Should Data download this book?" — the browser loop: SAPI David, rate 0, dry, silence trimmed, +5.0 dB with a −1.5 dBFS limiter; 1.5 s; −16.6 LUFS
`etvintriguing`	Data: "I find it most intriguing." (TNG "Haven"), YouTube `IljgQLbA-GY`, 56.85–59.00 s, +23.7 dB with a limiter; −16.3 LUFS
`etvcurious`	Data: "I am curious" (TNG "Genesis", where Spot has her kittens; cut before "as to why Spot…"), YouTube `IljgQLbA-GY`, 800.80–801.56 s: whisper (medium.en) hears "I am curious" from 800.80 once faded and encoded, "I'm curious" from 800.90; 20 ms fade in, 30 ms out, +14.9 dB with a −1.5 dBFS limiter; 0.76 s; −16.3 LUFS
`etvread`	Data: "0.68 seconds, sir. For an android, that is nearly an eternity." (Star Trek: First Contact), YouTube `5xPOZrdrU_8`, 1.35 s to the end, so "0.68" starts on the bar (whisper: "0" at 0.00 s, "eternity" ends 9.42 s), 20 ms fade in, 50 ms out, +23.0 dB with a −1.5 dBFS limiter; 10.1 s; −16.6 LUFS
`agletter`	The 26 letter names, `a.mp3`…`z.mp3` (`n` 0–25): SAPI Microsoft David, typed "A." … "Z.", rate +1 (W at +3), vocoded on c2 g2 c3 g3 (`--dry 1.5`), silence trimmed, loudnorm; 0.24–0.49 s each, peak about −2 dBFS. Alone, whisper misreads several (A as "Okay", D and T as blank); laid out a letter every 0.5 s as ALL GREEN spells them, `small.en` and `medium.en` both read A–M, N–Z, CRANE, DEPTH and QUICK correctly
`agfive`	"Five letters. Six tries." — David, rate 0, vocoded on c2 g2 c3 g3 (`--dry 0.6`); 2.54 s, −20.0 LUFS. Both models read it exactly
`agchoose`	"Choose the next fragment." — same settings; 1.29 s, −19.2 LUFS. Both read it exactly
`agpicks`	"J. E. V. doesn't write. J. E. V. picks." — David, rate −1, vocoded (`--dry 1.5`); 3.49 s, −17.1 LUFS. `small.en` "JEV doesn't write. JEV picks."; `medium.en` "J-E-V doesn't write. J-E-V picks." Typed "Jevv", `small.en` heard Jebb and Jeb every time
`agend`	"End the response now." — typed "Ennnd the response now.", rate −3, vocoded (`--dry 2.5`); 1.55 s, −20.3 LUFS. Both read it exactly; typed "End", `small.en` heard "And" every time
`agcost`	"Under a tenth of a cent, a game." — rate 0, vocoded (`--dry 0.6`). Both read it exactly

All are mono MP3s at 22.05 kHz (48 kbps for the plain voices, 64 kbps for the rest). Vocoded lines are made with speak then vocode — see tools/vocals/.