{ "cells": [ { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from phonemizer import phonemize\n", "\n", "phonemes = phonemize(text, language='en-us', backend='espeak')\n", "print(phonemes)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "import nltk\n", "nltk.download('averaged_perceptron_tagger_eng')" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "from g2p_en import G2p\n", "g2p = G2p()\n", "print(g2p(\"banana\"))\n", "\n", "text = \"Pronunciation respelling systems for English have been developed primarily for use in dictionaries.\"\n", "text = \"\"\"\n", "I ain't sayin' that I always sleep alone\n", "I ain't sayin' that I ain't met no one else\n", "Done a little bit of midnight movin' on\n", "And I ain't sayin' when I do that it don't help\n", "Leave with someone, I don't know her name\n", "Pull her real close, but I leave some space\n", "And I could forget you, girl, for good\n", "Even though I should\n", "\"\"\"\n", "\n", "symbols = g2p(text)\n", "symbols = [symbol.strip(\"0\").strip(\"1\").strip(\"2\") for symbol in symbols]\n", "print(symbols)" ] }, { "cell_type": "code", "execution_count": 73, "metadata": {}, "outputs": [], "source": [ "import random\n", "import re\n", "from g2p_en import G2p\n", "\n", "# ARPAbet → dictionary-style respelling map\n", "ARPABET_TO_RESPELLING = {\n", " \"AA\": \"ah\", # father\n", " \"AE\": \"a\", # cat\n", " \"AH\": \"uh\", # about\n", " \"AO\": \"aw\", # caught\n", " \"AW\": \"ow\", # now\n", " \"AY\": \"eye\", # my\n", " \"B\": \"b\",\n", " \"CH\": \"ch\",\n", " \"D\": \"d\",\n", " \"DH\": \"th\", # this\n", " \"EH\": \"eh\", # bed\n", " \"ER\": \"ur\", # bird\n", " \"EY\": \"ay\", # say\n", " \"F\": \"f\",\n", " \"G\": \"g\",\n", " \"HH\": \"h\",\n", " \"IH\": \"i\", # bit\n", " \"IY\": \"ee\", # seat\n", " \"JH\": \"j\", # jam\n", " \"K\": \"k\",\n", " \"L\": \"l\",\n", " \"M\": \"m\",\n", " \"N\": \"n\",\n", " \"NG\": \"ng\",\n", " \"OW\": \"oh\", # go\n", " \"OY\": \"oy\", # boy\n", " \"P\": \"p\",\n", " \"R\": \"r\",\n", " \"S\": \"s\",\n", " \"SH\": \"sh\",\n", " \"T\": \"t\",\n", " \"TH\": \"th\", # thin\n", " \"UH\": \"oo\", # book (best guess for most readers)\n", " \"UW\": \"oo\", # food\n", " \"V\": \"v\",\n", " \"W\": \"w\",\n", " \"Y\": \"y\",\n", " \"Z\": \"z\",\n", " \"ZH\": \"zh\" # genre — if confusing, could use \"zher\"\n", "}\n", "\n", "COMMON_MISPRONOUNCED_RESPELLINGS = {\n", " # Foods & Ingredients\n", " \"acai\": \"ahsighee\",\n", " \"quinoa\": \"keenwah\",\n", " \"gnocchi\": \"nyokee\",\n", " \"bruschetta\": \"broosketta\",\n", " \"jalapeno\": \"halapainyo\",\n", " \"croissant\": \"cwahsahnt\",\n", " \"tortilla\": \"torteeya\",\n", " \"hors d'oeuvre\": \"orderv\",\n", " \"pho\": \"fuh\",\n", " \"gyro\": \"yeeroh\",\n", " \"paella\": \"pah-ehya\", # consider \"pahaya\"\n", "\n", " # Foreign loanwords (general)\n", " \"rendezvous\": \"rahndayvoo\",\n", " \"faux\": \"foe\",\n", " \"bourgeois\": \"boorzhwah\",\n", " \"genre\": \"zhahnruh\",\n", " \"liaison\": \"leeayzon\",\n", " \"debacle\": \"daybahkul\",\n", " \"ennui\": \"onwee\",\n", " \"niche\": \"neesh\",\n", " \"coup\": \"koo\",\n", " \"déjà vu\": \"dayzhahvoo\",\n", " \"au revoir\": \"oh ruhvwah\",\n", " \"carte blanche\": \"cart blawnsh\",\n", " \"fiancé\": \"feeahnsay\",\n", " \"résumé\": \"rezuhmay\",\n", "\n", " # Place names\n", " \"worcestershire\": \"wustersheer\",\n", " \"gloucester\": \"gloster\",\n", " \"leicester\": \"lester\",\n", " \"edinburgh\": \"edinburrah\",\n", " \"new orleans\": \"nawlins\",\n", " \"thames\": \"tems\",\n", " \"cairo\": \"kairo\",\n", " \"dover\": \"dohver\",\n", " \"reykjavik\": \"raykyaveek\",\n", " \"melbourne\": \"melbun\",\n", "\n", " # People & brands\n", " \"nike\": \"nikey\",\n", " \"adidas\": \"ahdeedas\",\n", " \"colonel\": \"kernel\",\n", " \"hermes\": \"airmez\",\n", " \"versace\": \"versahchee\",\n", " \"mozart\": \"mohzart\",\n", " \"goethe\": \"gurtuh\",\n", " \"chloe\": \"kloh-ee\",\n", " \"saoirse\": \"seer-shuh\",\n", " \"aoife\": \"ee-fuh\",\n", " \"siobhan\": \"shivawn\",\n", " \"rachel\": \"raychel\",\n", "\n", " # Silent letters\n", " \"bologna\": \"baloney\",\n", " \"debris\": \"duhbree\",\n", " \"salmon\": \"samon\",\n", " \"solder\": \"sodder\",\n", " \"aisle\": \"ile\",\n", " \"island\": \"iland\",\n", " \"plumber\": \"plummer\",\n", " \"subtle\": \"suttle\",\n", " \"receipt\": \"reseet\",\n", " \"honest\": \"onest\",\n", " \"heir\": \"air\",\n", " \"gnome\": \"nome\",\n", " \"knife\": \"nife\",\n", "\n", " # Other weird ones\n", " \"segue\": \"segway\",\n", " \"epitome\": \"uhpituhmee\",\n", " \"victuals\": \"vittles\",\n", " \"cache\": \"cash\",\n", " \"nausea\": \"nawzhuh\",\n", " \"gesture\": \"jeshchur\",\n", " \"nuclear\": \"nookyular\",\n", " \"pecan\": \"puhkahn\", # regional variant\n", " \"affluent\": \"aflooent\",\n", " \"et cetera\": \"etseteruh\",\n", " \"often\": \"offen\", # silent 't' variant\n", " \"almond\": \"ahmuhnd\", # regional\n", " \"clothes\": \"cloze\",\n", " \"wolf\": \"woof\",\n", " \"jewelry\": \"joolree\",\n", " \"Wednesday\": \"wensday\",\n", " \"February\": \"febyuary\",\n", " \"library\": \"librerry\",\n", " \"espresso\": \"espresso\", # commonly mispronounced \"expresso\"\n", " \"pronunciation\": \"pruhnunseeayshun\",\n", "\n", " # mroe \n", " \"anemone\": \"uhnemonee\",\n", " \"asterisk\": \"asterisk\", # often mispronounced as \"asterix\"\n", " \"balloon\": \"baloon\",\n", " \"banal\": \"buhnahl\", # or \"baynul\", varies\n", " \"baroque\": \"buhroke\",\n", " \"beignet\": \"benyay\",\n", " \"biopic\": \"biopic\", # often misread as \"bio-pick\", but \"bye-oh-pick\"\n", " \"celtic\": \"keltic\", # for sports teams\n", " \"chassis\": \"chassee\",\n", " \"chiaroscuro\": \"keeahroskuro\",\n", " \"coup de grâce\": \"koo duh grahss\",\n", " \"dais\": \"dayis\",\n", " \"demesne\": \"demeen\",\n", " \"duodenum\": \"dooahdeenum\", # varies\n", " \"echelon\": \"eshuhlon\",\n", " \"eider\": \"ider\",\n", " \"ensign\": \"ensun\",\n", " \"eschew\": \"eshchoo\",\n", " \"facetious\": \"fuhsee shus\",\n", " \"forte\": \"fortay\", # debated but common usage\n", " \"gaffe\": \"gaf\",\n", " \"gif\": \"jif\", # controversial but relevant\n", " \"grotesque\": \"grotessk\",\n", " \"guillotine\": \"geeuhteen\",\n", " \"halcyon\": \"halseeun\",\n", " \"harass\": \"huh rass\",\n", " \"hegemony\": \"hehjuhmonee\",\n", " \"hirsute\": \"hersoot\",\n", " \"hyperbole\": \"hyperbuhlee\",\n", " \"indict\": \"indite\",\n", " \"inquiry\": \"inquiree\",\n", " \"leisure\": \"leezhur\",\n", " \"lingerie\": \"lahnzhuray\",\n", " \"macabre\": \"muhkahbruh\",\n", " \"mauve\": \"moav\",\n", " \"melee\": \"maylay\",\n", " \"minutiae\": \"minoosheeay\",\n", " \"mischief\": \"misschif\",\n", " \"moi\": \"mwah\",\n", " \"nauseous\": \"nawshus\",\n", " \"neither\": \"nee ther\",\n", " \"omen\": \"ohmen\",\n", " \"onomatopoeia\": \"onuhmatuhpeeuh\",\n", " \"paradigm\": \"paradime\",\n", " \"phlegm\": \"flem\",\n", " \"precocious\": \"pruhcoshus\",\n", " \"prescient\": \"preshunt\",\n", " \"prologue\": \"prolawg\",\n", " \"pseudonym\": \"soodunim\",\n", " \"ptarmigan\": \"tarmigan\",\n", " \"queue\": \"kyoo\",\n", " \"rapport\": \"rapor\",\n", " \"reconnaissance\": \"rekonuhsuns\",\n", " \"reprise\": \"reprize\",\n", " \"reveille\": \"revuhlee\",\n", " \"rhetoric\": \"retuhric\",\n", " \"risqué\": \"riskay\",\n", " \"rout\": \"rowt\",\n", " \"sacrosanct\": \"sackrosankt\",\n", " \"sangfroid\": \"sangfwah\",\n", " \"schematic\": \"skeematic\",\n", " \"schism\": \"sizm\",\n", " \"sergeant\": \"sarjunt\",\n", " \"sinew\": \"sinyoo\",\n", " \"soiree\": \"swahray\",\n", " \"sovereign\": \"sovrun\",\n", " \"subtlety\": \"suttultee\",\n", " \"syrup\": \"sirup\",\n", " \"taut\": \"tawt\",\n", " \"timbre\": \"tambr\",\n", " \"tome\": \"tohm\",\n", " \"treatise\": \"tretis\",\n", " \"viscous\": \"viskus\",\n", " \"vogue\": \"vohg\",\n", " \"yacht\": \"yot\",\n", " \"zealot\": \"zellut\"\n", "}\n", "\n", "\n", "g2p = G2p()\n", "\n", "\n", "#def strip_stress(phoneme):\n", "# return re.sub(r'\\d', '', phoneme)\n", "\n", "def strip_stress(phoneme):\n", " return phoneme.strip(\"0\").strip(\"1\").strip(\"2\")\n", "\n", "def phonemes_to_respelling(phonemes):\n", " #stripped = [strip_stress(p) for p in phonemes if p.isalpha()]\n", " #print(stripped)\n", " stripped = [strip_stress(p) for p in phonemes]\n", " return ''.join(ARPABET_TO_RESPELLING.get(p, '') for p in stripped)\n", "\n", "def respell_random_words_in_text(text, N=3):\n", " text = text.replace(\"'\", \"\")\n", " # Tokenize words and punctuation separately\n", " words = re.findall(r'\\b\\w+\\b|\\W+', text)\n", " word_indices = [i for i, w in enumerate(words) if re.match(r'\\b\\w+\\b', w)]\n", " selected_indices = random.sample(word_indices, min(N, len(word_indices)))\n", "\n", " for i in selected_indices:\n", " word = words[i]\n", " # Check if the word is in our common mispronounced words dictionary\n", " if word.lower() in COMMON_MISPRONOUNCED_RESPELLINGS:\n", " respelling = COMMON_MISPRONOUNCED_RESPELLINGS[word.lower()]\n", " print(f\"{word} (from dictionary): {respelling}\")\n", " else:\n", " # If not in dictionary, use phoneme-based respelling\n", " phonemes = g2p(word)\n", " respelling = phonemes_to_respelling(phonemes)\n", " print(word, phonemes, respelling)\n", " \n", " if respelling:\n", " words[i] = respelling\n", "\n", " return ''.join(words)\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "text = \"\"\"\n", "colonel quinoa gnocchi Worcestershire mischievous niche rendezvous salmon affluent debris cache epitome segue bologna victuals acai chai solder pecan schedule\n", "\"\"\"\n", "\n", "text = \"\"\"\n", "I never let my heart go all the way\n", "Every time I try, I just hit the brakes\n", "And there's always a couple tryna take your place\n", "But I never fall in love, baby, just in case\n", "You wake up wantin' me out of the blue\n", "You lay down needin' what I used to do\n", "Yeah, every now and then, I go get a taste\n", "But I never fall in love, baby, just in case\n", "Just in case\n", "\"\"\"\n", "\n", "respelled_text = respell_random_words_in_text(text, N=10)\n", "print(respelled_text)\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# lets' write a function that takes a full text string\n", "# we will then split this into words (by spaces and newlines)\n", "# then we will apply g2p to a subset of random words (N)\n", "# then we will map the symbols to the respellings \n", "# then we will re-insert them into the text at the same positions\n", "# we will then return the full respelled text\n", "\n", "def respell_text(text, N=10):\n", " words = text.split(\" \")\n", " random_words = random.sample(words, N)\n", " random_words = [word for word in random_words if word != \"\"]\n", " " ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "string_respell = \"\"\n", "for symbol in symbols:\n", " if symbol in rapabet_to_respell:\n", " string_respell += rapabet_to_respell[symbol]\n", " else:\n", " string_respell += symbol\n", "print(string_respell)\n" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "import json\n", "\n", "ipa_to_respell = json.load(open(\"/home/christian/code/christian/metadata/phonemes/ipa_mappings_extended.json\"))\n", "\n" ] }, { "cell_type": "code", "execution_count": 4, "metadata": {}, "outputs": [], "source": [ "ipa_to_wikipedia = {\n", " \"tʃ\": \"ch\",\n", " \"ɡ\": \"gh\",\n", " \"h\": \"h\",\n", " \"hw\": \"wh\",\n", " \"dʒ\": \"j\",\n", " \"k\": \"k\",\n", " \"x\": \"kh\",\n", " \"ŋ\": \"ng\",\n", " \"s\": \"s\",\n", " \"ʃ\": \"sh\",\n", " \"θ\": \"th\",\n", " \"ð\": \"dh\",\n", " \"j\": \"y\",\n", " \"ʒ\": \"zh\",\n", " \"æ\": \"a\",\n", " \"eɪ\": \"ay\",\n", " \"ɛər\": \"air\",\n", " \"ɑː\": \"ah\",\n", " \"ɑːr\": \"ar\",\n", " \"ɛ\t\": \"e\",\n", " \"iː\t\": \"ee\",\n", " \"ɪər\": \"eer\",\n", " \"ɪ\": \"i\",\n", " \"aɪ\": \"y\",\n", " \"ɒ\": \"o\",\n", " \"oʊ\": \"oh\",\n", " \"ɔː\": \"aw\",\n", " \"ɔːr\": \"or\",\n", " \"ɔɪ\": \"oy\",\n", " \"ʊ\": \"uu\",\n", " \"ʊər\": \"oor\",\n", " \"uː\": \"oo\",\n", " \"aʊ\t\" : \"ow\",\n", " \"ʌ\" : \"u\",\n", " \"ɜːr\" : \"ur\",\n", " \"ə\" : \"ə\",\n", " \"ər\" : \"ər\",\n", " \"juː\" : \"ew\",\n", "}\n", "\n", "rapabet_to_respell = {\n", " \"AA\": \"ah\",\n", " \"AE\": \"a\",\n", " \"AH\": \"uh\",\n", " \"AO\": \"aw\",\n", " \"AW\": \"ow\",\n", " \"AY\": \"eye\",\n", " \"B\": \"b\",\n", " \"CH\": \"ch\",\n", " \"D\": \"d\",\n", " \"DH\": \"th\",\n", " \"EH\": \"eh\",\n", " \"ER\": \"er\",\n", " \"EY\": \"ay\",\n", " \"F\": \"f\",\n", " \"G\": \"g\",\n", " \"HH\": \"h\",\n", " \"IH\": \"ih\",\n", " \"IY\": \"ee\",\n", " \"JH\": \"j\",\n", " \"K\": \"k\",\n", " \"L\": \"l\",\n", " \"M\": \"m\",\n", " \"N\": \"n\",\n", " \"NG\": \"ng\",\n", " \"OW\": \"oh\",\n", " \"OY\": \"oy\",\n", " \"P\": \"p\",\n", " \"R\": \"r\",\n", " \"S\": \"s\",\n", " \"SH\": \"sh\",\n", " \"T\": \"t\",\n", " \"TH\": \"th\",\n", " \"UH\": \"oo\",\n", " \"UW\": \"oo\",\n", " \"V\": \"v\",\n", " \"W\": \"w\",\n", " \"Y\": \"y\",\n", " \"Z\": \"z\",\n", " \"ZH\": \"zh\"\n", "}\n" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "# convert phonemes to respell\n", "respell = \"\"\n", "for char, phoneme in zip(text, phonemes):\n", " if phoneme in ipa_to_wikipedia:\n", " respell += ipa_to_wikipedia[phoneme]\n", " else:\n", " respell += char\n", "\n", "# join to a string\n", "string_respell = \"\".join(respell)\n", "print(string_respell)" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [ "for phoneme in phonemes:\n", " print(phoneme)\n", " # check if phoneme is in ipa_to_respell\n", " if phoneme in ipa_to_wikipedia:\n", " print(ipa_to_wikipedia[phoneme])\n", " else:\n", " print(\"not found\")\n", " print()" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "kernelspec": { "display_name": "suno_env", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.9" } }, "nbformat": 4, "nbformat_minor": 2 }