In college I worked part-time in the university's finance office, and often I would have to hold an account number of 5 or more digits in short-term memory as I went from the computer to a filing cabinet. For example, I read a purchase request for account 45749, realized I needed a physical file, then walked over to the cabinet to begin looking for that account number. The numbers all looked pretty similar, so I would be trying to find this account among a group like {47459, 49557, 44759, 94570, 45537, etc.}. Most of the time I could remember the number without any issues, and the first and last digits were very hard to forget, but out in the middle things got harder. If I messed up, I had to go back to the computer and try again. A reasonable person might decide to write down the number on a sticky note or something, but I had an idea that sounded more fun.
I assigned phonetic values to each digit as though they were letters (this post uses IPA for the number pronunciations). Many of them are based on similarities in shape, as well as pre-existing associations from "calculator words" like "07734" for "hello". So I started off with these simple correspondences:
0 1 2 3 4 5 6 7 8 9
o i e h s l b
2 looks like "z", but I decided to use "t" for the word "two", and similarly I used "n" for "nine". 6 looks like a capital G, so I chose "g", but got tripped up a lot at the beginning since 9 looks like a lowercase "g". I eventually got used to it, though, so the original system looked like this:
number 0 1 2 3 4 5 6 7 8 9
consonant t h s g l b n
vowel o i e
This worked as a proof of concept. The account numbers could now be memorized as nonsense words (much easier for me to remember than strings of digit names). Here are the ones mentioned above:
45749 47459 49557 44759 94570 45537
hslhn hlhsn hnssl hhlsn nhslo hssel
While this was possible through liberal use of syllabic consonants, clearly I needed more vowels and some euphony. The system has undergone several changes since then, but I'll omit those and just present the current version.
number 0 1 2 3 4 5 6 7 8 9
consonant ʃ j d θ f s g l b n
vowel o i y e a u ɤ ɯ ø æ
prefers V n C V V n C C C C
I memorized the digits by counting: "ʃo çi dy θe fa , su gɤ lɯ bø næ".
As you can see, each digit is now not only either a consonant or a vowel, but can serve as both depending on what is most comfortable. Each has naturally developed a preference for being a consonant or a vowel (1 and 5 have no preference either way), and this generally tries to avoid vowels other than [a e i o u].
Short numbers should alternate vowels and consonants if possible, but any configuration that is comfortable is acceptable. For example, "4938" can be "fæðø" or "anem", and I find the latter to be much easier to pronounce. The account numbers now have these preferred pronunciations:
45749 47459 49557 44759 94570 45537
aslan alawn ansul falun naslo fuzel
There is some allophony as well, some of which is already seen above. All plosives are voiced by default but can also be voiceless, usually word-finally or when bordering a fricative. /b/ and /g/ can be their corresponding nasals, usually only word-finally. /d/ does not do this because it would interfere with /n/, so instead it can become [ɾ] intervocalically or word-finally. Fricatives are voiceless by default but can become voiced, usually intervocalically. Nasals and /l/ used to be able to be syllabic, but I have largely done away with this in favor of vowels. The high vowels can also act as semivowels in vowel sequences but nowhere else, so I omit [ɥ ɰ w] but include [j], which is the primary consonant phoneme for its digit unlike the other three. Sequences of two of the same digit can be a long vowel or geminated consonant, although I generally avoid this.
The full details of allophony:
0: /ʃ/
[ʒ] intervocalically
[ʃ] otherwise
0: /o/
[o ~ o̞ ~ ɔ] in free variation
1: /j/
[ç ~ ʝ] when bordering [i], voiced intervocalically and voiceless otherwise
[j] otherwise
1: /i/ - [i] always
2: /d/
[t] when bordering a voiceless consonant, usually a fricative, or sometimes word-finally
[ɾ] intervocalically or word-finally (preferred over [t])
[d] otherwise
2: /y/ - [y] always
3: /θ/
[ð] intervocalically
[θ] otherwise
3: /e/
[e ~ e̞ ~ ɛ] in free variation
4: /f/
[h] in free variation with [f], usually word-initially and rarer otherwise
[v] intervocalically
[f] otherwise
4: /a/ - [a] always
5: /s/
[z] intervocalically
[s] otherwise
5: /u/ - [u] always
6: /g/
[k] when bordering a voiceless consonant, usually a fricative, or sometimes word-finally
[ŋ] word-finally (preferred over [k])
[g] otherwise
6: /ɤ/
[ɤ ~ ɤ̞ ~ ʌ] in free variation
7: /l/
[l ~ ɫ] in free variation, probably conditioned by nearby vowels but I haven't paid much attention to my habits here
7: /ɯ/ - [ɯ] always
8: /b/
[p] when bordering a voiceless consonant, usually a fricative, or sometimes word-finally
[m] word-finally (preferred over [p])
[b] otherwise
8: /ø/
[ø ~ ø̞ ~ œ] in free variation
9: /n/ - [n] always (must avoid assimilation with consonants to remain distinct from /b/ and /g/, so usually forces neighboring numbers to be vowels)
9: /æ/ - [æ] always
There is probably a simpler or more intuitive system for this, and maybe this doesn't need to exist at all, but I enjoy converting numbers to words and vice versa:
Chicago = 016460 (aspiration not marked)
librarians = 74182321395
geri dönüyorsunuz ("y'all are returning" in Turkish) = 6321 289210255955
I'd like to make a conlang using these phonemes (i.e., treating them as 10 phonemes, not 20) so that paradigms could have forms sounding completely different depending on the affixes used. For example, suppose a stem "826" can be inflected by the prefixes ["1" "2" "3"]. The results would be:
1-826 = ibyŋ ~ ibɾɤ (preference nCCC, so one of 2 or 6 should be a vowel)
2-826 = døɾɤ (preference CCCC, so two of them should be vowels, and CVCV is preferred over VCVC or other configurations)
3-826 = ebyŋ (preference VCCC, so 2 should be a vowel since CV alternation is preferred)
Wouldn't it be super fun to have to learn that "døɾɤ" and "ebyŋ" are from the same stem, and that "ibyŋ" and "ibɾɤ" are the same word?? This is one of my favorite bad ideas for a language so far.
But to go back to the original purpose, I have legitimately found this method useful for memorizing long numbers in a short amount of time. Phone numbers, account numbers, ZIP codes, you name it. I once memorized someone's transit card number (16 digits in 4 groups of 4) using this, and I still can't get rid of that now-long-term memory.
Maybe someone else will find this useful?