SPA-Scots: Writing Glasgow So a Machine Can Sing It A developer has proposed SPA-Scots (Stable Phonetic Alphabet for Scots English), a multiscript phonetic spelling system designed to give AI music generators unambiguous pronunciation input for broad Central Belt Scots with a Glasgow lean. Version 1.1, the "Multiscript Build," maps one sound to one symbol, combining Latin letters with German diacritics, þ and ð, and single Georgian, Cyrillic and Greek characters to replace English digraphs such as sh, ch and ng. The system targets the ambiguous spellings, cross-linguistic confusion and under-represented training data that cause singing models to default toward English pronunciations. A multiscript phonetic spelling for Central Belt Scots — version 1.1 Standard Scots spelling is a poor input for AI music generators: dreich , scunner and wheesht are traditions, not instructions, and models tend to read them as English. The International Phonetic Alphabet IPA is precise, but it is hard to type and not reliably interpreted by singing models. This document proposes SPA-Scots Stable Phonetic Alphabet for Scots English as a practical, experimental alternative for broad Central Belt Scots with a Glasgow lean. SPA-Scots is a system for writing pronunciation, designed for AI singers. Its core principle is one sound = one symbol , and a symbol keeps its sound everywhere. This version v1.1, “Multiscript Build” uses Latin letters as a base, extended with German letters ä, ü, ö , þ and ð, and single characters from Georgian, Cyrillic and Greek for sounds that English spells with digraphs sh , ch , ng . Design rules: - Phonemic, not orthographic: write what is pronounced, not what the standard spelling says. - No silent letters. - Rhotic: keep every /r/ the speaker pronounces bird , car , hard . - Diphthongs are fixed two-letter units ai , oi , au . - One speaker, one variety: transcribe the actual accent, not an imagined universal Scots. - No decorative script mixing: every symbol has one defined value. AI music generators can be unreliable when faced with: - Ambiguous spellings: Scots ui , ou , ae and ch have no fixed value in English. - Cross-linguistic confusion: ch is /tʃ/ in church but /x/ in loch . - Under-represented varieties: little Scots in the training data means poor guesses, usually towards English. SPA-Scots aims to give the model a clear, consistent input. - Example 1: The ‘ch’ and ‘sh’ conflict - English ch = /tʃ/ church , chiel → SPA ჩ - Scots ch = /x/ dreich , loch → SPA ხ - English sh = /ʃ/ shite → SPA შ - English - Example 2: The ‘ou’ and ‘oo’ conflict - English ou = /aʊ/ house → SPA au - Scots oo = /ʉ/ hoose , oot → SPA ü - English - Example 3: The ‘wh’ conflict - English wh = /w/ which → SPA w - Scots wh = /ʍ/ whit , whaur → SPA ў - English | IPA | SPA | Source/Justification | |---|---|---| | Stops | | | | /p/ | p | Latin | | /b/ | b | Latin | | /t/ | t | Latin | | /d/ | d | Latin | | /k/ | k | Latin | | /g/ | g | Latin | | Fricatives | | | | /f/ | f | Latin | | /v/ | v | Latin | | /θ/ | þ | Icelandic, Old English, Old Norse | | /ð/ | ð | Icelandic, Old English, Old Norse | | /s/ | s | Latin | | /z/ | z | Latin | | /ʃ/ | შ | Georgian; one character instead of the digraph sh | | /ʒ/ | ჟ | Georgian | | /x/ | ხ | Georgian; loch , dreich | | /h/ | h | Latin | | Affricates | | | | /tʃ/ | ჩ | Georgian; one character instead of the digraph ch | | /dʒ/ | ჯ | Georgian | | Nasals & Approximants | | | | /m/ | m | Latin | | /n/ | n | Latin | | /ŋ/ | ң | Cyrillic Kazakh, Kyrgyz ; avoids the /n/+/g/ ambiguity of ng | | /l/ | l | Latin | | /r/ | r | Latin; always pronounced rhotic | | /j/ | y | Latin; unambiguous | | /w/ | w | Latin | | /ʍ/ | ў | Cyrillic Belarusian ; whit , whaur | | Vowels | | | | /iː/ | í | Diacritic convention tense | | /ɪ/ | ι | Greek iota; kept apart from í and from the diphthong glide i | | /e/ | é | Diacritic convention; Scots face and nae are monophthongs | | /ɛ/ | ä | German same value | | /a/ | a | Latin; trap | | /ɑː/ | â | Diacritic convention; braw , hard | | /ɒ/ | ó | Diacritic convention; lot | | /ɔː/ | ô | Diacritic convention; thought | | /ʌ/ | ă | Romanian Latin diacritic ; strut | | /ʊ/ | ù | Diacritic convention; foot in Scottish Standard speech | | /uː/ | ú | Diacritic convention; goose in Scottish Standard speech | | /ʉ/ | ü | German closest single character ; hoose , oot , and foot , goose , mouth in broad Scots | | /ø/ | ö | German; traditional guid optional | | /o/ | o | Latin; goat | | /ə/ | ə | IPA letter, also Azerbaijani; unstressed vowels | | Diphthongs | | | | /aɪ/ | ai | Latin digraph; price , aye | | /ɔɪ/ | oi | Latin digraph; choice | | /aʊ/ | au | Latin digraph; mouth in Scottish Standard speech only | Notes: - The letters i and u occur only as diphthong glides. - Scots NURSE words have no vowel of their own: write vowel + r bird → bärd . - ß is not used: Scots has no sound beyond /s/ for it. - Arabic خ ش ج exist only as an explicitly designated alternative variant; never mix them with the Georgian letters. | Scots | IPA | SPA-Scots | |---|---|---| | braw | /brɑː/ | brâ | | wee | /wiː/ | wí | | hoose | /hʉs/ | hüs | | oot | /ʉt/ | üt | | dreich | /driːx/ | dríხ | | wheesht | /ʍiːʃt/ | ўíშt | | shite | /ʃaɪt/ | შait | | ken | /kɛn/ | kän | | bairn | /bɛrn/ | bärn | | muckle | /ˈmʌkəl/ | măkəl | | scunner | /ˈskʌnər/ | skănər | | cannae | /ˈkane/ | kané | A full lyric line: “Big fancy words, but nae bloody plan ” → Bιg fansí wărdz, bət né blădí plan Paste this into an LLM to convert English or Scots lyrics into SPA-Scots. You are a transcription assistant. Convert English or Scots lyrics into SPA-Scots v1.1 Stable Phonetic Alphabet for Scots English : broad Central Belt, Glasgow-influenced, rhotic. Transcribe what is pronounced, not what is spelled. Use ONLY the symbols below. Never substitute look-alike characters Cyrillic х for ხ, Greek Ι for Latin I . The format is IPA - SPA . SPA-Scots Transcription System CONSONANTS Stops: /p/- p, /b/- b, /t/- t, /d/- d, /k/- k, /g/- g Fricatives: /f/- f, /v/- v, /θ/- þ, /ð/- ð, /s/- s, /z/- z, /ʃ/- შ, /ʒ/- ჟ, /x/- ხ, /h/- h Affricates: /tʃ/- ჩ, /dʒ/- ჯ Nasals: /m/- m, /n/- n, /ŋ/- ң Approximants: /l/- l, /r/- r always pronounced , /j/- y, /w/- w, /ʍ/- ў VOWELS /iː/- í; /ɪ/- ι; /e/- é; /ɛ/- ä; /a/- a; /ɑː/- â; /ɒ/- ó; /ɔː/- ô; /ʌ/- ă; /ʊ/- ù; /uː/- ú; /ʉ/- ü; /ø/- ö optional, traditional Scots only ; /o/- o; /ə/- ə DIPHTHONGS /aɪ/- ai; /ɔɪ/- oi; /aʊ/- au RULES No silent letters. Keep every /r/. There is no separate NURSE vowel: write vowel + r bird - bärd . FACE and GOAT are monophthongs: é and o day - dé, go - go . In broad Scots the FOOT, GOOSE and MOUTH vowels are all /ʉ/ - ü look - lük, about - əbüt . Final unstressed -y is í money - măní . No apostrophes. Use hyphens between syllables only where the source has them. Never use ß, θ or ng. Write /ŋ/ as ң. Lines that start with ι stay lowercase Greek capital Ι is identical to Latin I . Keep section tags unchanged. Output only the transcription. SPA-Scots is an evolving experiment. This document reports no model test results, so treat every choice above as a hypothesis until it has been tested. Known risks: - Non-Latin letters may pull the accent. The general SPA proposal SPA: A Stable Phonetic Alphabet for AI Music https://gist.github.com/MrDoe/6a9b13f3806df87d3f7d7c63330dc6db reported that too many Cyrillic letters made Suno sound like Russian. It switched to Latin digraphs, which performed better in its Swabian tests. SPA-Scots uses two Cyrillic and five Georgian letters, so test these first. - Model-dependent values: ä, ü and ι may be read differently by different models. A German-trained model should give ä as /ɛ/ and ü as a rounded close vowel; ι is Greek but looks like a Latin i . - Look-alike characters: the first full conversion a song found a Cyrillic х where the Georgian ხ belonged, plus a stray IPA ɪ, a Greek ε and θ, and a Latin æ. Scan the codepoints and normalise to NFC. - Dialect coverage: one accent only. There is no symbol yet for the glottal stop Glasgow speakers use for /t/ butter , water , and vowel values vary by region and speaker. Fallback build proposal, untested : if non-Latin letters misfire, swap them for Latin digraphs, following the general SPA v0.3 table. Never mix the two builds in one text. | SPA-Scots | Fallback | IPA | |---|---|---| | შ | sh | /ʃ/ | | ჟ | zh | /ʒ/ | | ჩ | ch | /tʃ/ | | ჯ | dj | /dʒ/ | | ხ | chh | /x/ | | ң | ngh | /ŋ/ | | ў | hw | /ʍ/ no precedent in the general table | Other fallbacks: ú instead of ü for foot and goose words, au in the mouth words, and ß for every /s/ if a model voices s between vowels. How to Help: - Test with AI: use SPA-Scots in AI music generators and note the model and version. Does it improve pronunciation? - Identify Issues: which symbols are sung wrongly or read as another language? Which Scots sounds are missing? - Suggest Improvements: propose better symbols, digraphs or conventions based on your tests. The general SPA proposal also suggested separate tables per base language; SPA-Scots is one such table. The goal is more predictable Scots vocals from AI singers, through tests anyone can repeat.