Community Bonding Period : 9th May to 1st June 2025

20th May'2025

• First Meet with Domenico + Nahuel
• Laid the foundation on how we gonna start working
  • Reciter first -> Parser in Phausto at a later stage in summer
  • Make API for PAM -> just like this one for PAM : LINK
  • Lot's of mini test cases
  • a pair programming sess in JUNE -> to see how Nahuel uses Pharo adequatly

To Do's for me :

• SAM repo go through + specifcially the Reciter part
• connect a Github repo to iceberg -> share with Domenico + Nahuel as contributors
• start with SAM reciter translation in Pharo -> add it in the same above repo

22nd May'2025

• Started with the self study of SAM

  • 
  • 
  • 

• SAM uses 450 rules to convert English into SAM’s phonetic language

• Phoneme Rules of SAM :

  • 

• Fetch SAM api sam-js from npm

  • code at SAM API Fetch
  • here's output :
    • 

• Insights I got to know

  • Audio dosent start on its on browser -> we need a user interaction to start it
  • 
  • hence we have a button to play it + event listeners (for user interaction )

23rd May'2025

• trying to understand the reciter -> constants.es6 file + reciter -> tables.es6

• code the Phonemes in SAM.pdf book .. and make a visulaization of it

• code file at : text

  • output at :

Insights I got to know

• why is there a quote export const FLAG_ALPHA_OR_QUOT = 0x80; in reciter -> constant.es6

• example :

  • in the word "banana" /bə'nænə/, the quote before 'n' indicates that the second syllable 'na' is stressed.

• use of charFlags in reciter -> constants.es6

  1. 0x01 - Numeric Flag

  • Example: Digits 0-9
  • Usage:
    • Used to identify numbers in the input text
    • Example: In "I have 2 apples", '2' would have this flag set

  2. 0x02 - Rule Set 2 Flag

  • Example: Most punctuation (!, ", #, $, etc.)
  • Usage:
    • Special processing rules for punctuation and symbols
    • Example: In "Hello!", the '!' would use a different rule set

  3. 0x04 - Voiced Consonants (D, J, L, N, R, S, T, Z)

  • Example: 'D' in "dog", 'J' in "jump"
  • Usage:
    • Identifies consonants where vocal cords vibrate
    • Helps in determining pronunciation rules

  4. 0x08 - Special Consonants (B, D, G, J, L, M, N, R, V, W, Z)

  • Example: 'M' in "mother", 'B' in "baby"
  • Usage:
    • Used for specific consonant processing
    • Basic consonanats

  5. 0x10 - Fricatives/Affricates (C, G, J, S, X, Z)

  • Example: 'S' in "sun", 'CH' in "church"
  • Usage:
    • Identifies fricative and affricate sounds
    • Important for proper pronunciation of hissing/hushing sounds

  6. 0x20 - Consonant Flag

  • Example: All consonants (A-Z except vowels+ Y)
  • Usage:
    • General flag for any consonant
    • Example: 'C' in "cat", 'K' in "kite"

  7. 0x40 - Vowel or Y

  • Example: 'A' in "apple", 'Y' in "happy"
  • Usage:
    • Identifies vowels and the sometimes-vowel 'Y'
    • Important for syllable counting and stress rules

  8. 0x80 - Alphabetic or Quote

  • Example: 'A'-'Z', 'a'-'z', and single quote (')
  • Usage:
    • General flag for letters and stress markers
      Example: All letters in "can't" would have this flag, plus the quote

• Example :

• How are individual charFlags checcke?

  • you add them with the respective FLAG_CONSTANT from the reciter -> constants.es6
  • 

24th May'2025

• Symbols used in reciter -> tables.es6

  • understanding FORCED VOWELS
    • u speak bringing tongue in fwd position
    • see common exmaples below
    • 

• from export const rules = ""

• some exmpales :

  • #:(ALS) = ULZ| -> if begining of word (#:) is ALS -> replace with ULZ when saying

    • eg. ALSO -> ULSo

  • '#:(AG)E=IHJ| -> if in begining of word (#:) is AG foloowed by E-> replace with IHJ when saying

    • eg. age -> IHJe
    • pronouced as :
      • IH: Like the 'i' in "bit" or "sit" (short 'i' sound)
      • J: Like the 'y' in "yes" or the 'j' in "jump"

  • '(A)^+#=EY5|' -> a A not follwed by end ^+# should be EY

    • eg. make -> mEYk

  • '(AR)#=EH4R -> AR at end

    • eg. BURGLAR -> buhglEHR

  • '#:(ALLY)=ULIY|' -> ally at end should be ULIY

    • eg. magically -? magiCULIY

  • '(AL)^=AOL|' -> al not followed by anything ->ie al at the end of sentence

    • eg. bacterial ->bacteriAOL

  • ' (BUS)#=BIH4Z|'+ -> bus at the begining should be BIH4Z

    • eg. Businnes -> BIH4Zinnes

  • (CH)^=K -> ch at end .. not followed by anything

    • eg. stomach -> stomaK

  • ' S(CI)#=SAY4|'+ -> S at begining followed by CI ... at the end ... should be pronouced say

    • eg. SCI FI -> SAYFI

  • '#:&(ES) =IHZ|' -> es anywhere in the middle ... (#: ... second half of the word) should be ihz

    • eg. fresh -> frIHZh
    • eg. guess -> guIHZh

  • '#:R(IED) =IYD|' -> in the second half .. after ( #: ) -> if there is a RIED .. IYD

    • eg. cried -> crIYD

  • '#^(ON)=UN|' -> start of the word is not ON .. then on is un

    • eg. onion -> oniUN
    • eg on is not the matching the rule ... hence pronounced OON

  • '#(SION)=ZHUN|' -> after start if there is a SION ... make it ZHUN

    • eg. vision -> viZHUN

  • '@(UR)#=UH4R|' -> if ur anywhere .. make it voiced consonant

    • eg. burial -> biUH4Rl

• some words to rule translation

• COLLEGE == KOW5ULEHG

• SCHOOL == SKUW5UL

• from export const rules2 = ""
• some examples :
  • has all pronouciations , numbers
  • '(") =-AH5NKWOWT-|' -> a single double quote sounds like : "a-n-kwote"
  • (^)= KAE4RIXT -> caret

25th May'2025

English words -> Phoneme Conversion

• understanding the reciterRule in reciter.es6
• Getting pre , match , post target

  const splitted = ruleString.split('=');
  const
    // Must pop and join here because of rule for '=' itself.
    target = splitted.pop(),
    source = splitted.join('=').split('('),
    tmp=source.pop().split(')'),
    pre = source[0],
    match = tmp[0],
    post = tmp[1]

• Eg.
  1. In '#:(ALL)Y=ULIY|'
     • target = ULIY|
     • pre = #:
     • match = ALL
     • post = Y
  2. In ANS(WER) = ER|
     • target = ER|
     • pre = ANS
     • match = WER
     • post = ""

understandign checkPrefix function in reciter.es6

• for every current letter in English word ->
  • First -> finds matching rule for each letter
  • second -> checks backwards from the letter ... to see if any rule matches
  • the net pronouceiation == that which complies with both FIRST + SECOND
  • 
  • 

understanding the TextToPhoneme function in reciter.es6

• make a .env and have the property DEBUG_SAM = true to seee magic in console 😊
• see the letter which is being processed -> in RED CAPS
• 

    if (process.env.DEBUG_SAM === true) {
      const tmp = text.toLowerCase();
      console.log(
        `processing "${tmp.substr(0, inputPos)}%c${tmp[inputPos].toUpperCase()}%c${tmp.substr(inputPos + 1)}"`,
        'color: red;',
        'color:normal;'
      );
    }

understand the pre reqs of the reciter.es6

// Map all rules and generate processors from them.
const rules = {};
tables.rules.split('|').map((rule) => {
  const r = reciterRule(rule), c= r.c;
  rules[c] = rules[c] || [];
  rules[c].push(r);
});
const rules2 = tables.rules2.split('|').map(reciterRule);

• make a rules dict having a dictionary of all rules
  • key : alphabets
  • value : all rules for the aplbhabet

understand closures in JS

• 
  • after the outer function finishes executing ... const myInnerFunc = outerFunc();
  • when the innerFunc is called -> it returns its lexical scope -> ie. inner function itself + outer variable
  • closure is the act of returning of lexical scope

function backpackMaker() {
  let snack = "cookie"; // snack packed in the backpack

  function showSnack() {
    console.log("Snack inside the backpack is: " + snack);
  }

  return showSnack;
}

let myBackpack = backpackMaker(); // You get the backpack from school
myBackpack(); // You open the backpack at home

// output : Snack inside the backpack is: cookie

• eg. You are in school, and your teacher gives you a backpack with some snacks 🍪 inside. You go home, and the school is closed. But you still have your backpack with the snacks inside! 🎒

• Even though the school (where you got the backpack) is closed, you can still eat your snacks because the backpack remembers what was inside it.

understand function flow in reciter.es6

1. result() -> matches() -> match() + checkPrefix() + checkSuffix()

  result.c = match[0];
  return result

• at this point for every rule going in reciterRule() -> we have pre , match , post , target calculated
• we have defined function matches , checkPrefix , checkSuffix , result -> so tjey have closures on pre, post , match , target
• result.c -> match[0] ; this result is returned
• we have rules / rules2 dict ... having all rules ... it is dictinorised with key == match[0]
• eg.

rules = {
  'A': [/* rule for A */],
  'B': [/* rule for B */],
  'C': [/* rule for CH */],  // Note: 'CH' is bucketed under 'C'
  'T': [/* rule for T in TION */]
}

understand flags funciton

// matching case -> true 
// non matching case -> false
const flags = (c, flg) => {
  return (tables.charFlags[c] & flg) !== 0;
};

• if both are matching then not equal to 0 --> itll be equal to FLAG value
• eg.
  • DIAPHTHONG FLAG = 0x08 and G == Diaphthong
  • G -> 0x80 | 0x20 | 0x10 | 0x08
  • G = 1011 0010
  • G & DIAPHTHONG FLAG = 1011 0010 & 0000 1010 != 0
  • hence in matching case -> true

26th May'2025

understand .some() function in js

• checks if atleast one element in the array satisfies the condition
• 

let fruits = ["apple", "banana", "mango", "carrot"];

// arrow function
fruits.some((fruit) => fruit === "banana"); // true

// normal function
let hasMango = fruits.some(function(fruit) {
  return fruit === "mango";
});

console.log(hasMango); // true 

understand flow of TextToPhoneme

export const TextToPhonemes = (input) => {

// output -> FINAL RULE STRING 
  let inputPos = 0, output = '';

  // will have rule string + and pointer of next round ... it shall increment every time
  const successCallback = (append, inputSkip) => {
    inputPos += inputSkip;
    output += append;
  };

  while ((inputPos < text.length) && (c++ < 10000)) {
    const currentChar = text[inputPos];

    // NOT '.' or '.' followed by number.

    // this if -> for rules2 ( punctuation +  number )
    // ! followed by number -> decimals ( they have normal processing )
    if ((currentChar !== '.')
      || (flagsAt(text, inputPos + 1, FLAG_NUMERIC))) {

      if (flags(currentChar, FLAG_RULESET2)) {
        rules2.some((rule) => {
          return rule(text, inputPos, successCallback);
        });

        continue;
      }

      // tables.charFlags == 0 where ( , ) , [ , ] , " " , _ ,

      // for all alphabets + quote
      if (tables.charFlags[currentChar] !== 0) {
        if (!flags(currentChar, FLAG_ALPHA_OR_QUOT)) {
          return false;
        }
        rules[currentChar].some((rule) => {
          return rule(text, inputPos, successCallback);
        });
        continue;
      }

      // for any random shit not defined in tables
      // this is not adding after every rule ...  we see continue after everything every succesfull match ( rules2 , rule ) n above two if blocks

      output += ' ';
      inputPos++;
      continue;
    }

    // for the end wala fullstop -> is not a part of first if block
    // the pronouciation does not sound anything
    output += '.';
    inputPos++;
  }
  return output;
};

• eg.
  • 
  • 
  • 
  • 
  • 
  • 

decoding the myth behind result()

• is result a object or a function based on the below code snippet ?

  const result = (text, inputPos, callback) => {
    if (matches(text, inputPos)) {
      if (process.env.DEBUG_SAM === true) {
        console.log(`${source} -> ${target}`)
      }
      callback(target, match.length);
      return true;
    }
  };
  result.c = match[0];

  return result;

• 
• hence,
  • when we do result.c -> its a object
  • when we do result() -> its a function (function call has parentheses)

understanding wht is returend in the line const r = reciterRule(rule), c= r.c; ... result as a fucntion or result as a object

const rules = {};
tables.rules.split('|').map((rule) => {
  const r = reciterRule(rule), c= r.c;
  rules[c] = rules[c] || [];
  rules[c].push(r);
});

understand function flow in reciter.es6

1. rules dict/rules2 dict -> (pre + match + post + target) + result function having closures of (matches + checkPrefix + checkSuffix) + result obj having c property ( signifying the match[0] character )

2. TextToPhonemes function -> define inputPos + output string + checks if valid alphabet ( calls rule1 -> result() function ) + checks if valid number ( calls rules2 -> result() function ) + increment pos + append output

Coding Peiod officially starts !

1st June'2025

Added repo to github using Iceberg

• Repo Link : PAM-GSoC-25-Project

• added tutorial to my ongoing Pharo tutorial repo

  • Link : Learn-With-Me-Pharo

in later stages of project instead of Dictionary add a Trie to make it faster

Revised Pharo again ! 😊

7th June'2025

Hehe need to buckle up fast now 😅

• revised the handwritten notebook notes
• revised the readme.md written notes ... so know everythign now 😇

understanding [options.phonetic] in index.es6

• options.phonetic defines that wether our input string ( one we put in text box of SAM) is already phonemene string or not
• if it is phonemene string (HE2LL5OW) -> return it in caps
• else convert the text ( normal english Hello) to its phonemene string

how to change it to true .. can be done when making a new instance of it

understanding class-side vs instance-side in pharo

• a brief :

  • 

  • class side is equivalent to static methods

    • 

  • instance side is equivalent to defining methods on the objects only ( not on class directly)

    • 

THUMB RULE : as CONSTANTS are static -> should be on class side !!

• Example of class side method and how to call it :

  • 

• Example of instance side method and how to call it:

  • 

Some other way to declare class-instance variable

• instance variable
  • 
• class variable
  • 

FIRST ACHIEVEMENT : ❤️

• Commit Link : PAM-GSoC-25-Project

• added reciter constants in pharo along with its tests

• 

8th June'2025

conversion of tables constants to pharo

• Javascript Version to refer and make pharo code ! 😊

  • code at : Mapping eng to decimal values
  • output :
    

• pharo version !! 😇

  • 
  • done with tests too !!!
    • 

11th June'2025

• Meet with Naheul and Domenico
  • Discussed about all work done by me till now -> they are happy !
  • saw my written readme.md
  • focus on completing the below test snippet

  testCollegePhoneme

	| phonemeOutput inputText  |
	inputText := 'COLLEGE'.
	
	phonemeOutput := reciter textToPhoneme: inputText.
  "reciter == a object of class that deals with actual text to phoneme conversion"

	self assert: phonemeOutput equals: 'KOW5ULEHG'.

13th June'2025

understanding Pharo Method structure :

understanding how we are trying to build this thing of Rules

• structurally

  • 
  • 

• example

  • *

• visual representation

  • 
  • 

achievement :

• sucessfully converted the alphabets rules to PHARO

• added the CRUD ops on it as well

  • 

• tom -> all rest alphabets + tests + puctuation

15th June'2025

understanding class side variables and class side instance variable and instance variable

• a java to pharo comparison

• instance variable => each obj has one copy [ same for java + [pharo]]
  • 

- class variables => class has its copy [ in java its called static ]

• 

• similar to public static int age;

• coz of public -> all its subclass can have the same copy of the age variable

• class side instance variable => the class where it is defined only has the copy .. not shared in subclasses

  • 
  • similar to private static int age;
  • coz of private -> no subclass can use it
  • a secret safe place only for the class it is defined in !!

• 

• 

y we use class side instance varibales if static can be used?

• its a safe secret place for the class
• some uses are :
  • 
  • 

understanding classPool

• Every class can have class variables (shared among all instances of the class and its subclasses).

• Pharo stores these varibles in they're kept in a special dictionary called the classPool.

• so you can get the key - value pair of class Variables

• code to acess the class variables via class pool

"syntax:"
"ClassName classpool at: classVariableName"

| rules |

rules := A classPool at: #RulesDictionary. 

Transcript show: 'Access class Variables via class pool'.

Transcript show: (A classPool at: #RulesDictionary) size printString; cr.

rules do: [ :rule |
    Transcript
        show: rule pattern;
        show: ' =>';
        show: rule replacement;
        cr
].

Transcript open.

some mistakes while using classvariable

• class A defintion -> it has a class variable RulesDictionary which we can access via class pool

• also definition of initializing method

"class definition"

LetterRules << #A
	slots: {};
	sharedVariables: { #RulesDictionary };
	tag: 'Rules';
	package: 'PAM-Core'

"initliazation method"
initialiseRules
    "Initialize rules for letter A"
    RulesDictionary := OrderedCollection new.

    "Add all A rules"
    self addRule: ' (A.)' replacement: 'EH4Y.'.
    self addRule: '(A) ' replacement: 'AH'.

• error :
• solution :

"hence our intiliazation method becomes "

initialiseRules
    "Initialize rules for letter A"
    self classPool at: #RulesDictionary put: OrderedCollection new.

    "Add all A rules"
    self addRule: ' (A.)' replacement: 'EH4Y.'.
    self addRule: '(A) ' replacement: 'AH'.

how to create class and subclass in Pharo

• class creation :

Object subclass: #Rules
    instanceVariableNames: 'pattern replacement'
    classVariableNames: ''
    poolDictionaries: ''
    category: 'PAM-Core-Rules'.

• subclass creation :

Rules subclass: #AdvancedRules
    instanceVariableNames: 'priority'
    classVariableNames: ''
    poolDictionaries: ''
    category: 'PAM-Core-Rules'.

• checking is subclass or not ?

AdvancedRules superclass.   "Returns: Rules"

The current pharo uses ShiftClassBuilder format to define classes

setup and tearDown methods

achivements :

• refactored code [ like names of classes + variables]

• parent and child relation achieved .. things make better sense now

  • added doubts on discord + asked Nahuel.. as wanted aall methods to be in parent class .. only intialisation method in child

• test case added [ also achived a functionality by setUp and tearDown .. that the orignal Dictionary dosent get disturbed when running testcases !]

• replied to Domenico on mail

• pic of the day 😊
  

17th June'2025

• receieved a reply on discord + github issue + mail reply highlighting coding practises

• here's a ref : Top down approach

• here's a gpt of wht is expected of me .. did not understand at first 😢

  • 
  • 
  • 
  • 

• so well i need to do a top down approach -> think about the prob at large ... and solve from the point of view of a user .. not the coder

• meaning .. not how to code [ this is wht i did wrong .. i did constants and tables code etc etc ] ... then did the oops level thing [ static , parent , child .. etc etc]

the question of how to code -> has many answers .. it will bring in many design principles

in the path of how to answer thethe above -> we loose the wht the issue was .. the user problem

do not do that

hence try tdd , take a step back and analyze the exact problem before falling in this trap

• 

• with that said .. this is what Nahue expects :

  • 
  • 
  • 

18th June'25

• meet with Nahuel and Domenico -> recieved certain feedback :

1. Remove unnecessary loggers from test.
2. Add a transcript of steps of text to phoneme conversion when writing reciter code.
3. change word 'replacement' to 'phoneme' in the methods.
4. add a 'matchFor: pattern' in the methods

23rd June'25

doing a pharo code conversion of reciter.es6 logic :

finding prefix pattern suffix of every rule's LHS :

• pharo index starts at 1

• 

• so for rule : #:^(I)^#

  • $( : 4
  • $) : 6
  • length of string : 8
  • prefix : #:^ [ 1 to 3 ]
  • pattern : I [ 5 to 5 ]
  • suffix : ^# [ 7 to 8 ]

• 

• output :

checking of isSpecialCharacter

isSpecialCharacter: char at: position in: text
    "Check if the character should be treated specially"
    | flags |
    flags := Flags flagsFor: char.
    "Check if character is not alphabetic or quote"
    (flags anyMask: 128) not  "FLAG_ALPHA_OR_QUOT"
        ifTrue: [ ^ true ].
    
    "Check for period not followed by a number"
    (char = $. and: [ 
        (position + 1 > text size) or: 
            [ ((Flags flagsFor: (text at: position + 1)) anyMask: 1) not ]  "FLAG_NUMERIC"
    ]) ifTrue: [ ^ true ].
    
    ^ false

• so in rule processing ... special characters [ alphabets , fullstop at end .. etc have impt]

• 

• use of anymask a utility helper to check and: things :

• 

24th June'25

• achievement of the day : added rules for HELLO word + code addition of textToPhoneme method

• 

• lets see if it works 😉

28th June'25

in @reciter.es6 , wht is happening ? are all rules being checked for a character and best one is returened ? or only yht frirst best is found and returned ?

• what i did -> check for all rules -> which fit the pattern in input text we are trying to replace
• then return the longest pattern match

  • like if 2 rules match (B)->B and (BREAK) -> BRE5K

  • return BREAK !!

  • why so ?

    • the shorter one will match the pattern first as in matched in BREAK irrespective
    • but the longer one will be the correct one which needs to be matched

kindaa i can say ... the multiple parts of textToPhoneme function work

understading our matchSpecialCharacter: char in: text at: position direction: dir method

first the JS version with examples :

• 

• 

• 

• 

• 

• each of these special characters like @ # ^ have same logic -> its just the direction of search is different

// in checkSuffix 

// '#' - next char must be a vowel or Y.
'#': () => flagsAt(text, ++pos, FLAG_VOWEL_OR_Y),

// in checkPrefix 

// '#' - previous char must be a vowel or Y.
'#': () => flagsAt(text, --pos, FLAG_VOWEL_OR_Y),

textToPhoneme works words of A and B !! ❤️

• a quick meet with Nahuel helped in denugging the code : therefore the textToPhoneme method ran !

  • 

• wrote a green and yellow text too !!! ❤️

  • 
  • yellow coz rules for that word need to be implemented as well

• discussion in meet

  • debug why rules for H E L O aint comming in Dictionary

    • ( thanks to Nahuel i could understand that my rules parsing was not correct -> was adding a = while intilising rules)
    • ^ output was causing th error in textToPhoneme -> thanks agaian Nahuel

  • make these two test yellow

  • post that we need to stop reciter -> than focus on Phausto implementation ( parser at a later stage !)

why do we need to add space at strat of word ?

• as seen here .. every letter which is at start needs a space to denote it is a starting letter word ...
• 
• hence in smalltalk 'textToPhoeneme' we add space to start of word

textToPhonemes: input
    "Convert input text to phonemes"
    | text output pos charFlags rule currentChar |
    text := '  ',input asUppercase. "Pad with space to justify start of a word"
    output := ''.
    pos := 1.

    "rest of code .."

achievement of the day 1:

• `textToPhoeneme worksssss
• 
• 

achievement of the day 2:

let's make a logger like Springboot's Sl4j

log: message

    | senderContext className methodName |
    
    "Get the sender context (the method that called log:) -> so the parent method inside which log: is called is picked up from here"
    senderContext := thisContext sender.
    className := senderContext method methodClass name.
    methodName := senderContext selector.
    
    Transcript
        nextPutAll: '[';
        nextPutAll: className;
        nextPutAll: ' >> #';
        nextPutAll: methodName asString;
        nextPutAll: '] ';
        nextPutAll: message;
        cr.
    Transcript flush

• debugmode is a class variable which is initally false -> hence nothing seen in transcript

  • 

• let's make it true !

  • 
  • 

removed leading and trailing spaces from output

"via this line in textToPhoneme at the end before returnign output"

"Remove leading and trailing spaces"
output := output trimBoth.

29th June'25

Quick meet with Domenico on using TpSamplee and playing numbers:

• numbers folder -> put in pharo image

• in playground -> find the location of folder :
  pathToFolder := FileLocator image asAbsolute pathString , '/numbers'.

• use TpSampler : is a sample player ( sample : piece of recorded music .. like number folder )

• pathToNumbers := FileLocator imageDirectory asAbsolute pathString , '/numbers'.

• samplePlayer := TpSampler new pathToFolder: pathToNumbers

• ctrl+B -> shortcut to go in browser of class

• folder for number then next folder for alpabets

• a simple tests we need to achieve

pathToNumbers := FileLocator  imageDirectory asAbsolute pathString , '/numbers'. 

samplePlayer := TpSampler new pathToFolder: pathToNumbers .

dsp := samplePlayer stereo asDsp.

| samplePlayer |

self deny: samplePlayer stereo asDsp isNull.

• make a tag inside Pam-Core : instrument
• name as PamNumberSampler
• PAMDSP -> classSide : sayZeroSample ,

study pragma in pharo

• code snippets taken from the meet

• adding label

• dsp trace allParams
  

• triggering the buttons with code
  

• pragma
  

• <script> makes test like playing ( run the class ) in class side
  

sayZeroExample
<script>
pathToNumbers := FileLocator  imageDirectory asAbsolute pathString , '/numbers'. 

samplePlayer := TpSampler new label: 'Numbers'; pathToFolder: pathToNumbers .

dsp := samplePlayer stereo asDsp.
dsp init.
dsp start.
dsp setValue: 0 parameter: 'NumbersIndex'.
dsp trig: 'NumbersGate'.

• learn process in pharo ( multithread )

  • see if timeRepeat is a process

• 

• 

  • use fork

• use fork with delay in time

  • 
  • delay between in each phoenme ( h -> 1sec later e -> 1 sec later l .. etc)

• play all numbers in a line

  • 

• goal for this week : run code line at 29

  • 
  • smth like this :

"THIS IS BADDDDDDDD , BUT LOOK AT THE DEBUGGER ! myString := '1 2 3 4'."

[  myString do:  [ : i |
		dsp setValue: i  parameter: 'NumbersIndex'.
       dsp trig: 'NumbersGate'.
  
		. 1 seconds wait ] ] fork

• endgoal 2 : take a string : from sampler play audio
  

"you can make another example: "

sayOneToFour
myString := '1 2 3 4'.

[  myString do:  [ : i |
		dsp setValue: (myString indexOf: i ) parameter: 'NumbersIndex'.
       dsp trig: 'NumbersGate'.
  
		. 1 seconds wait ] ] fork

important book to understand phonetic sounds -> will come handy !!

Sounds of Spoken English : internet link

Sounds of Spoken English : downloaded

5th July'25

Achievement of the day : 😇

• added all letter rules
• tried a few conversions -> working fine ( needs some refining --> but this is good to start !!!! )

6th July'25

• 🌻

Achievement of the day --> completed all tests for all words

• textToPhoneme works fine ❤️

• sentence also works fimne ... but ... space consideration needs to happen --> WILL do in next phase !

Starting a fresh -> downloading Phausto to start with the reciter

differnce in mono and stereo and bringing out audio !!:

• 

• the code worked .. u need to select the entire thing for it to work .. doing it line by line .. will throw error 😢

• but here's the code whihc worked

  • 

• some displayUI magicccc

  • 

• adding labels

  • 

• triggering buttons without the UI

  • learning wht each control button is called :

    • 

  • triggereing them .. chaging their values

    • 

• playing numbers audio

1. all in a 0 to 9 format -> we need to fix 7 , 8 , 9 .. as they are not audible 😢

2. playing the numbers in a predefined list format

• made their respective methods !!
• entire code !!!

| pathToNumbers samplePlayer dsp |

pathToNumbers := FileLocator imageDirectory asAbsolute pathString , '\numbers' "'C:\Users\Neerja\Documents\Pharo\images\PAM v1\numbers'". 
samplePlayer := TpSampler new label: 'NumberAudio' ; pathToFolder: pathToNumbers. "a TpSampler"

dsp := samplePlayer stereo asDsp. "a DSP(@ 16r1BA1931ECC0)"

dsp init. "a DSP(@ 16r1BA1931AD60)"

dsp start. "true"

dsp traceAllParams. "a DSP(@ 16r1BA2A25D4F0)"

dsp setValue: 5 parameter: 'NumberAudioIndex'. "a DSP(@16r1BA2A25E6E0)"

dsp trig: 'NumberAudioGate'. "a DSP(@ 16r1BA2A25EC00)"

dsp displayUI "a BlSpace('MyApp UI')".

dsp stop. "a DSP(@ 16r1BA1AE11D30)"

dsp destroy. "a DSP(@ 16r1BA23FB4D70)"


"playing all numbers"
(0 to: 9) do: [:i |
    dsp setValue: i parameter: 'NumberAudioIndex'.
    dsp trig: 'NumberAudioGate'.
    (Delay forSeconds: 1) wait.  "Pause 1 second between numbers"
].

"playing in a predefined list manner"
#(5 3 1 9 0) do: [:i |
    dsp setValue: i parameter: 'NumberAudioIndex'.
    dsp trig: 'NumberAudioGate'.
    (Delay forSeconds: 1) wait.
].

andddd best part pf the day !!! 😍

• made methods to do this

  • 

  • the best commit !! : first DSP commit

  • also opened a issue side by side in phausto -> coz it was error causing when i was coding about it ---> Phausto Issue

7th July'25

• asked doubt about pragmas on Discord ... turns out they are just annotations like @service in springboot

• 

9th July '25 -- a quick call with Domenico

• why test fails:
  

• setup method .. to intilizise everythign ..

• make a new image -> pull the repo -> to check wht is missing .. where are the inconsistenecies.

• pragma error resolve

  • 

  • 

• issue opened in discord :

  • 

• corrcet this :

• 

• 

AlphabetUtility initializeDictionary.
Reciter textToPhonemes: 'music'.
Domenico Cipriani
14:32
playAllNumbersWithDelay
	"plays all numbers in a linear fashion"

	[(0 to: 9) do: [ :i |
			dsp setValue: i parameter: 'NumberIndex'.
			dsp trig: 'NumberGate'.
			(Delay forSeconds: 1) wait ] ] fork

• mac vs window folder issue

• eval feedback :

• slow at first -> but new to pharo so understable

• on pace now

• have some not working code then no code at all

• do reciter complete ( tests intilizations + instarument bugs ) -> move to parser

• from parser we need duration and stress

19th July

learned how to take pull in a new image :

• add the ssh clone link from repo

  • 
  • this shall clone the repo ... but but ...
  • it shall give u package not loaded error
  • 

• fix for package not loaded : METHOD 1

  • u need to load packages individually

  • 

  • 

  • now packages are up to date :

  • 

• fix for package not loaded : METHOD 2

  • we need to have a baseline Metacello ... need to with Domenico / Nahuel for this
  • see this vid at time stamp 12:30
    • Iceberg Clone Link at 12:30
  • Metacello is like a package manager for Pharo (just like pip , npm .. loads packaged automatically)
  • so u need to have a Package Metacello baseline !!

  "A initialiseRules."
"A printRules."
"A getAllRules."

LetterRules initialiseRules. "LetterRules"
N getAllRules.
N addRule: 'NN' replacement: 'NENE'. 
Reciter textToPhonemes: 'NN'. "'NENE'"

M printRules.

M classPool .
M initialiseRules.
M printRules.
M printRules.

LetterRules removeRule: 'M' replacement: 'EH4M'. 

Transcript open.
AlphabetUtility initializeDictionary. 
AlphabetUtility rulesFor: $M.


Reciter textToPhonemes: 'nAHUEL'. "'NAE/HYUWEHL'"

Reciter textToPhonemes: 'MONA'. "'EH4MOWNAE'"

"og:"
Reciter textToPhonemes: 'Domenico'. "'DOWMEHNICAA'"
Reciter textToPhonemes: 'Magnificent'. "'MEY5GNIFICEHNT'"
Reciter textToPhonemes: 'Emirates'. "'EHMIRAETEHS'"

"new:"
Reciter textToPhonemes: 'Domenico'. "'DUWMEHNIHKOW'"
Reciter textToPhonemes: 'Magnificent'. "'EH4MEY5JNIHFIHKEHNT'"
Reciter textToPhonemes: 'Emirates'. "'EHMAYRAETEHS'"

making synchronous calls asynchronous

• wait vs fork

• 

• 

• the wait used to block pharo -- cant do anything else, not even type

• ▶ Watch demo

• Live demo :

• hence we use, fork :

• 

• ▶ Watch demo

• Live demo :
  

refactorings in instrument class

• so now we dont need to hardcode label as in this commit ...
• have made a genralzed method --> and the folder specific methods calls the genralized method .. setting parameters

"generalized"
initilizeWithLabel: aLabel folderName: aFolderName 
	"a generalised method to set label and folder Name in Windows environment !"
    | path |
    path := FileLocator imageDirectory asAbsolute pathString ,  , aFolderName.
    
    samplePlayer := TpSampler new 
        label: aLabel; 
        pathToFolder: path.

    dsp := samplePlayer stereo asDsp.
    dsp init.
    dsp start.

"folder specific methods"

initializeNumbersFolder
	"pre requisite --> have the numbers audion folder in image directory"
	self initilizeWithLabel: 'Number' folderName: 'numbers'.

• so this is how it works now :

|mydsp player|

"old method -- forget this .. just for ref"
mydsp:= PAMDsp new.
mydsp playAllNumbersWithDelay.
mydsp playNumbers: #( 0 0 7 8 6).
mydsp cleanup.

"new method .. use this !"
player := PAMDsp new.
player initializeNumbersFolder.
player playAllNumbersWithDelay.
player playNumbers: #(1 3 5 7).

quick meet with Domenico

• make an array of this: (aka replacement for us)

• 

• the big picture

• 

• for me to pass the eval .. make the thing speak -- use replacement from here 😂

• 

• windows macos path discussion

OSPlatform current isMacOSX
OSPlatform current isWindows

quick notepad notes

• in every mp3 file --> the amplitude wave is of -1 to 1

• [ they render a array of amplyude values .. 0.4 , 0.5 , 0.55 .. ]

• we doing concatenative synthesis -> add phonemen audio together

diphones = [
    'IY', 'IH', 'EH', 'AE', 'AA', 'AH', 'AO', 'UH', 'AX', 'IX',
    'ER', 'UX', 'OH', 'RX', 'LX', 'WX', 'YX', 'WH', 'R*', 'L*',
    'W*', 'Y*', 'M*', 'N*', 'NX', 'DX', 'Q*', 'S*', 'SH', 'F*',
    'TH', '/H', '/X', 'Z*', 'ZH', 'V*', 'DH', 'CH', 'J*',
    'EY', 'AY', 'OY', 'AW', 'OW', 'UW', 'B*', 'D*', 'G*', 'GX',
    'P*', 'T*', 'K*', 'KX', 'UL', 'UM', 'UN'
]

"make an array of this: "
Reciter textToArrayOfPhonemes" aString

Reciter textToArrayOfPhonemes: 'celsius'
^ #('*' 's*' 'eh' 'lx' 's*' 'ih' 'ah' 's*')

• write a test .. taht covered all 47 phonemen
• textToArrayOfPhonemeTest ---> cover all words from textToPhoneme Tests

todos

• make array of replacements ( arrayOfPhoneme )
• add numbers folder in repo
• add readme !!

correct version of making pragmas -- learnt from Domenico!!

playAllNumbersWithDelay
"pragma for the above method"
  
	<script: 'PAMDsp new playAllNumbersWithDelay'> " --- incorrect .. no need of adding class-methodname"
	<script: 'self playAllNumbersWithDelay'> " --- correct .. add self methodname"
	
	|player|	
	player := PAMDsp new.
	player initializeNumbersFolder.
	player playAllNumbersWithDelay.

windows and Mac folder issue solved !!

• undersatnding path separators

• 

• therefore in this commit .. added pathSeparator logic

• commit link

two major bugs fixed !! --> acheivement of the day ❤️

1) reason why sentence parsing wasn't correct

• in textToPhoneme --> when we want to skip characters .. we should not add the pos
  • 
  • 

2) reason why we used to get wrong phonemes for certain words

• exmaple :

• in the rules dictinary of every letter --> we used to have simple rule .. ie addRule: '(G)' replacement: 'G' in the end

• we also have another rule addRule: ' (G)' replacement: 'JIY4' .. ie. start of word rule

• as we select the rule with longest pattern --> we do not overwrite --> hence, we used to have this rule in the above image: addRule: ' (G)' replacement: 'JIY4'

• becaus of this --> sunflower had smth like the image above

• hence we do 2 things --

  • 1st

    • change the order of rules in the resp POJO
    • 
    • so from next repo pull onwards --> same mistake not repeated

  • 2nd

    • in the current image .... remove and add rule such that ... the the order is maiantained .. hence .. correct rule matching
    • 
    • see the below code for having the order mainatained

G removeRule: ' (G)' replacement: 'JIY4'.
G removeRule: '(G)' replacement: 'G'.
G addRule: '(G)' replacement: 'G'.
G addRule: ' (G)' replacement: 'JIY4'.
G getAllRules.


F removeRule: ' (F)' replacement: 'EH4F'.
F removeRule: '(F)' replacement: 'F'.
F addRule: '(F)' replacement: 'F'.
F addRule: ' (F)' replacement: 'EH4F'.
F getAllRules.

J removeRule: ' (J)' replacement: 'JEY4'.
J removeRule: '(J)' replacement: 'J'.
J addRule: '(J)' replacement: 'J'.
J addRule: ' (J)' replacement: 'JEY4'.
J getAllRules.

K removeRule: ' (K)' replacement: 'KEY4'.
K removeRule: '(K)' replacement: 'K'.
K addRule: '(K)' replacement: 'K'.
K addRule: ' (K)' replacement: 'KEY4'.
K getAllRules.

K removeRule: ' (K)' replacement: 'KEY4'.
K removeRule: '(K)' replacement: 'K'.
K addRule: '(K)' replacement: 'K'.
K addRule: ' (K)' replacement: 'KEY4'.
K getAllRules.

L removeRule: ' (L)' replacement: 'EH4L'.
L removeRule: '(L)' replacement: 'L'.
L addRule: '(L)' replacement: 'L'.
L addRule: ' (L)' replacement: 'EH4L'.
L getAllRules.

M removeRule: ' (M)' replacement: 'EH4M'.
M removeRule: '(M)' replacement: 'M'.
M addRule: '(M)' replacement: 'M'.
M addRule: ' (M)' replacement: 'EH4M'.
M getAllRules.

N removeRule: ' (N)' replacement: 'EH4N'.
N removeRule: '(N)' replacement: 'N'.
N addRule: '(N)' replacement: 'N'.
N addRule: ' (N)' replacement: 'EH4N'.
N getAllRules.

P removeRule: ' (P)' replacement: 'PIY4'.
P removeRule: '(P)' replacement: 'P'.
P addRule: '(P)' replacement: 'P'.
P addRule: ' (P)' replacement: 'PIY4'.
P getAllRules.

R removeRule: ' (R)' replacement: 'AA5R'.
R removeRule: '(R)' replacement: 'R'.
R addRule: '(R)' replacement: 'R'.
R addRule: ' (R)' replacement: 'AA5R'.
R getAllRules.

S removeRule: ' (S)' replacement: 'EH4S'.
S removeRule: '(S)' replacement: 'S'.
S addRule: '(S)' replacement: 'S'.
S addRule: ' (S)' replacement: 'EH4S'.
S getAllRules.

T removeRule: ' (T)' replacement: 'TIY4'.
T removeRule: '(T)' replacement: 'T'.
T addRule: '(T)' replacement: 'T'.
T addRule: ' (T)' replacement: 'TIY4'.
T getAllRules.

V removeRule: ' (V)' replacement: 'VIY4'.
V removeRule: '(V)' replacement: 'V'.
V addRule: '(V)' replacement: 'V'.
V addRule: ' (V)' replacement: 'VIY4'.
V getAllRules.

W removeRule: ' (W)' replacement: 'DAH4BULYUW'.
W removeRule: '(W)' replacement: 'W'.
W addRule: '(W)' replacement: 'W'.
W addRule: ' (W)' replacement: 'DAH4BULYUW'.
W getAllRules.

some more happy news today !!

• GSOC midterm eval passed !! --> worthwhile feedback received from mentors 😊

  • 

• first half of stipend received !! 😃

  • 

30th July '25

Understading the different ways in which we have phoneme representation

• ARPABet : capital letters ( SAM uses this )

• IPA : greek letters

• Links to understand :

  • IPA with Example words

  • ARPA to IPA table + python coversion code of ARPABet to IPA

  • dataset of 44 phonemes :

    • Link

code to get the ARPA format audio samples from SAM

const fs = require('fs');
const path = require('path');
const { Sam } = require('./sam'); // Adjust path if necessary

const phonemes = [
  "AA", "AE", "AH", "AO", "AW", "AY",
  "B", "CH", "D", "DH", "EH", "ER",
  "EY", "F", "G", "HH", "IH", "IY",
  "JH", "K", "L", "M", "N", "NG",
  "OW", "OY", "P", "R", "S", "SH",
  "T", "TH", "UH", "UW", "V", "W",
  "Y", "Z", "ZH"
];

// Create output directory
const outDir = path.join(__dirname, 'phonemes_wav');
if (!fs.existsSync(outDir)) fs.mkdirSync(outDir);

// Function to synthesize and save
phonemes.forEach(ph => {
  const sam = new Sam();
  sam.sayPhonemes(ph); // Core phoneme input
  const wavBuffer = sam.toWav(); // Get WAV as buffer
  fs.writeFileSync(path.join(outDir, `${ph}.wav`), wavBuffer);
  console.log(`Saved ${ph}.wav`);
});

// to run 
// node generatePhonemeAudio.js

difference in what Domenico gave ( Diphones ) vs what we are downloading ( Phoneme )

// Domenico's list
diphones = [
    'IY', 'IH', 'EH', 'AE', 'AA', 'AH', 'AO', 'UH', 'AX', 'IX',
    'ER', 'UX', 'OH', 'RX', 'LX', 'WX', 'YX', 'WH', 'R*', 'L*',
    'W*', 'Y*', 'M*', 'N*', 'NX', 'DX', 'Q*', 'S*', 'SH', 'F*',
    'TH', '/H', '/X', 'Z*', 'ZH', 'V*', 'DH', 'CH', 'J*',
    'EY', 'AY', 'OY', 'AW', 'OW', 'UW', 'B*', 'D*', 'G*', 'GX',
    'P*', 'T*', 'K*', 'KX', 'UL', 'UM', 'UN'
]

// my list
phonemes = [
  "AA", "AE", "AH", "AO", "AW", "AY",
  "B", "CH", "D", "DH", "EH", "ER",
  "EY", "F", "G", "HH", "IH", "IY",
  "JH", "K", "L", "M", "N", "NG",
  "OW", "OY", "P", "R", "S", "SH",
  "T", "TH", "UH", "UW", "V", "W",
  "Y", "Z", "ZH"
];

• Phonemes -- standalone :

  • 
  • 

• Diphones -- the blend of second half of first phoneme and first half of second phoneme :

  • 
  • 

• in terms of count of both :

  • 

• diffference in / and * in diphones :

  • 

  • 

  • example :

    • 
    • 

grapheme ( the way we write sounds in simple eng )

actual conversion

• the audio files we have are having these phoneme :
  • 
  • they have named some as colon
  • like this :

Code changes

• Reciter textToPhoneme method now returns array of phoenems .. insyead of a string

• 

• we then convert the phoneme array to IPA format ( we will connect audio via this !! 😗)

steps for phoneme --> IPA conversion ( as per filenames)

we used recursion and greedyyy approachhhhh 😭😎❤️

"code to run"
|PAMPhonemeString|

PAMPhonemeString:= Reciter textToPhonemes: 'Czech Republic.'.
PAMPhonemeString. "#('S' 'Z' 'EH' 'K' 'RIY' 'P' 'AH5' 'B' 'L' 'IH' 'K')" 

PhonemeToIPAConvertor initializeDictionary.
PhonemeToIPAConvertor convertPAMPhonemeToIPAFormat: PAMPhonemeString.  

"#('s' 'z' 'e' 'k' 'r' 'i_colon' 'p' 'ʌ' 'b' 'l' 'i' 'k')"

• strip the phoentic string of any digits ( these are our stress markers .. we dont need them here )

• make a dictionary -> where we store basic phoneme to IPA

  • 

• for every phoneme we got from textToPhoneme method --> check if a direct conversion is available in dictionary

  • if yes --> we got our IPA

  • if no --> we have compound Phoneme RIY --> R + IY

  • for them we use recursion --> greedy approach ( match the longest possible pattern available in dictionary .. recurse to small then)

splitCompoundPhoneme: PAMPhoneme
	"Split compound phoneme into known phonemes using a greedy approach --> ie longest first"
	| stripped size result pos found |
	stripped := self stripStressMarkersInPhoneme: PAMPhoneme.
	size := stripped size.
	result := OrderedCollection new.
	pos := 1.

	[ pos <= size ] whileTrue: [
		found := false.
		(size to: pos by: -1) do: [ :end |
			| slice |
			slice := (stripped copyFrom: pos to: end).
			(IPAConversionDictionary includesKey: slice) ifTrue: [
				result add: slice.
				pos := end + 1.
				found := true.
				^ result , (self splitCompoundPhoneme: (stripped copyFrom: pos to: size)) "recurse on the rest"
			]
		].
		found ifFalse: [
			result add: (stripped copyFrom: pos to: pos).
			pos := pos + 1.
		]
	].

	^ result

• exmaples of abpve recursion code
  • 
  • 

Todo

• add these github badges on pharo project readme
  • badge link

creating our project baseline :

dependencies loaded ?

so our metacello script now becomes !!! 😎

Metacello new
    baseline: 'PAM';
    repository: 'github://neerja-1984/PAM-GSoC-25-Project:master/src';
    load.

let's try loading metacello in a new image !!

• error ( metacello conflict )
  • 

sucessfully added numbers audio files in github repo

4th Aug : Quick Meet with Domenico

reason why code not working --> reason for this bug

• need to Modify TPSmapler --> as now only WideString (will not consider IPA characters )
• hence doemico added ByteString instead of Widestring --> need to pull Phausto iamge again !

"do this again :"
Metacello new
 baseline: 'CoypuIDE';
 repository: 'github://pharo-graphics/CoypuIDE/';
 onConflict: [ :ex | ex useIncoming ];
 onUpgrade: [ :ex | ex useIncoming ];
 load

 "later do a git pull"

some other error

1. metadata issue:

• audio files to run in phausto --> should not have metadata
• hence : https://online-audio-converter.com/ --> removes metadata from files

2. libsndfile issue:

• https://en.wikipedia.org/wiki/Libsndfile

• It seems like you're referring to libsndfile, a library used for reading and writing sound files. If libsndfile is not reading files with metadata correctly, there could be several reasons for this issue. Here are some steps you can take to troubleshoot and potentially resolve the problem:

• Check the File Format: Ensure that the file format you are trying to read is supported by libsndfile. Some formats might not support metadata or might require special handling.
  the libdynamic-engine.dylib that Phausto is using is linked to libsndfile to read audio files

• we can fix this issue only by fixing libsndfile hard work with C!

instead of a new class --> make this in Reciter textToIPAPhonemes method

textToIPAPhonemes: aString

	^ PhonemeToIPAConvertor convertPAMPhonemeToIPAFormat:
		  (self textToPhonemes: aString)

a small bug fix i need to do

giftPhonemes := Reciter textToPhonemes: 'gift'.
#('G' 'IH' 'F' 'T')

giftIPAString := Reciter textToIPAPhonemes: 'gift'.
^ #('g' 'i' 'f' 't')

• hence map IH --> i_colon

• 

• 

what are we targetting by gsoc end ?

todo

• add all tests from testToPhoneme -> textToIPAPhonemes
• https://en.wiktionary.org/wiki/gift : for checking how the IPA phoneme sounds
• add audio folder in documents folder path rather than image
• textToIPAPhonemes --> in Reciter class
• see the bugs in testToPhoneme -->
• add the phoneme_dataset
• pull phausto

code snippets Domenico shared

"in PAMDSP class; "
phonemesFolder
	^ FileLocator documents asAbsolute pathString , '/phonemes_dataset' -

sampler := TpSampler new pathToFolder: PAMDsp phonemesFolder; label: 'PAMSampler
sampler sampleNamesSorted

dsp := sampler stereo asDsp.
dsp init.
dsp start.
myString := Reciter textToIPAPhonemes: 'hello'.
[myString do: [ : i | ( list indexOf: i ) traceCr. dsp setValue: ( list  indexOf: i)  parameter: 'PAMSamplerIndex' . dsp trig: 'PAMSamplerGate'  . 0.2 seconds wait] ] fork.


[myString do: [ : i |  dsp setValue: ( list  indexOf: i)  parameter: 'PAMSamplerIndex' . dsp trig: 'PAMSamplerGate'  . 0.2 seconds wait] ] fork.

pam := PAMDsp new.
pam say: 'hello'.
say: aString

sortedListOfPhonemes := self sortedListOfPhonemes.
[(Reciter textToIPAPhonemes: aString ) do: [ : i |  self setValue: ( sortedListOfPhonemes  indexOf: i)  parameter: 'PAMSamplerIndex' . self trig: 'PAMSamplerGate'  . 0.2 seconds wait] ] fork.

#('?' '?' 'e' 'l' 'l' 'əʊ')
dsp setValue: 261 parameter: 'PAMSamplerNote'.
this last method goes aftger the dsp initialization

PhonemeToIPAConvertor initializeDictionary. this can go inside PAMDsp initialization

5th August'25 --- starting with rework

how to print Documents absolutepath :

• 
• henceforth, audio samples can be present in Documents path itself --> instead of having it in iamge

adding extra checks to get only audio files ( no hidden files )

x := FileLocator documents asAbsolute pathString , '\phoneme_dataset'.
x createMultiFilesURL.

x asFileReference children reject: [ :f | f isHidden ].
x asFileReference children collect: [ :f | f isAudioFile ].

why audio samples should be sorted ?

x := FileLocator documents asAbsolute pathString , '\phoneme_dataset'. "'C:\Users\Neerja\Documents\phonemes_dataset'"

sampler := TpSampler new pathToFolder: x ; label: 'PAMSampler'.
sampler sampleNamesSorted.

• what does the code of sampleSirted give us ? : files soerted alphabetically
• why we need sorted files ->
  • phausto will mark each in synchronous order aɪ:1 ; b:2 .. so on
  • hence, on picking phonemes audio --> it will map the IPAphoneme --> to its number --> get that numbered file form folder
  • eg : HU -> u_colon -> 12 file in folder

 "#(
  '''C:\Users\Neerja\Documents\phoneme_dataset\a_colon' '''C:\Users\Neerja\Documents\phoneme_dataset\aɪ' '''C:\Users\Neerja\Documents\phoneme_dataset\aʊ' '''C:\Users\Neerja\Documents\phoneme_dataset\b' '''C:\Users\Neerja\Documents\phoneme_dataset\d' '''C:\Users\Neerja\Documents\phoneme_dataset\dʒ' '''C:\Users\Neerja\Documents\phoneme_dataset\e' '''C:\Users\Neerja\Documents\phoneme_dataset\eə' '''C:\Users\Neerja\Documents\phoneme_dataset\eɪ' '''C:\Users\Neerja\Documents\phoneme_dataset\f' '''C:\Users\Neerja\Documents\phoneme_dataset\g' '''C:\Users\Neerja\Documents\phoneme_dataset\h' '''C:\Users\Neerja\Documents\phoneme_dataset\i' '''C:\Users\Neerja\Documents\phoneme_dataset\i_colon' '''C:\Users\Neerja\Documents\phoneme_dataset\j' '''C:\Users\Neerja\Documents\phoneme_dataset\k' '''C:\Users\Neerja\Documents\phoneme_dataset\l' '''C:\Users\Neerja\Documents\phoneme_dataset\m' '''C:\Users\Neerja\Documents\phoneme_dataset\n' '''C:\Users\Neerja\Documents\phoneme_dataset\p' '''C:\Users\Neerja\Documents\phoneme_dataset\r' '''C:\Users\Neerja\Documents\phoneme_dataset\s' '''C:\Users\Neerja\Documents\phoneme_dataset\t' '''C:\Users\Neerja\Documents\phoneme_dataset\u_colon' '''C:\Users\Neerja\Documents\phoneme_dataset\v' '''C:\Users\Neerja\Documents\phoneme_dataset\w' '''C:\Users\Neerja\Documents\phoneme_dataset\z' '''C:\Users\Neerja\Documents\phoneme_dataset\æ' '''C:\Users\Neerja\Documents\phoneme_dataset\ð' '''C:\Users\Neerja\Documents\phoneme_dataset\ŋ' '''C:\Users\Neerja\Documents\phoneme_dataset\ɒ' '''C:\Users\Neerja\Documents\phoneme_dataset\ɔ_colon' '''C:\Users\Neerja\Documents\phoneme_dataset\ɔɪ' '''C:\Users\Neerja\Documents\phoneme_dataset\ə' '''C:\Users\Neerja\Documents\phoneme_dataset\əʊ' '''C:\Users\Neerja\Documents\phoneme_dataset\ɜ_colon' '''C:\Users\Neerja\Documents\phoneme_dataset\ɪə' '''C:\Users\Neerja\Documents\phoneme_dataset\ʃ' '''C:\Users\Neerja\Documents\phoneme_dataset\ʈʃ' '''C:\Users\Neerja\Documents\phoneme_dataset\ʊ' '''C:\Users\Neerja\Documents\phoneme_dataset\ʊə' '''C:\Users\Neerja\Documents\phoneme_dataset\ʌ' '''C:\Users\Neerja\Documents\phoneme_dataset\ʒ' '''C:\Users\Neerja\Documents\phoneme_dataset\θ'
  )"

• what is the pattern of our files in folder ?

why sortign is impt ?

• phausto understands numbers --> not strings
• so for our TPSampler aka phausto to make sound work --> we need to provide indexes of the files / IPAPhoneme we want to run

| path pathSeparator x sampler musicString list dsp musicIPAString myString ipaList ipaListSorted|

"1. making the sampler"
x := FileLocator documents asAbsolute pathString , '\phoneme_dataset'.  "'C:\Users\Neerja\Documents\phoneme_dataset'"
sampler := TpSampler new pathToFolder: x ; label: 'PAMSampler'. "a TpSampler"

"wht is sorted list"
list := sampler sampleNamesSorted.

"does not retrun sorted list"
ipaList := (x asFileReference children collect: [ :f | f basenameWithoutExtension ]).

"returns sorted list"
ipaListSorted := ipaList asSortedCollection asArray.

• 

• testing :

10th August'25

HOw the whole things works

• Give a folder -> make a TPSampler of it -> make a DSP of it -> set the Index param ( the file index we want to run ) -. trig the Gate param to play the audio

• 

• 

• 

where and how should we provide sorting ?

1. LIke the ipaListSorted --> always works :
2. TPSampler --> sampleNamesSorted --> never works :

What's the real issue with Domenico's code.

|  path pathSeparator x sampler musicString list dsp musicIPAString myString ipaList ipaListSorted|

"1. making the sampler"
x := FileLocator documents asAbsolute pathString , '\phoneme_dataset'.  "'C:\Users\Neerja\Documents\phoneme_dataset'"
sampler := TpSampler new pathToFolder: x ; label: 'PAMSampler'. "a TpSampler"

"wht is sorted list"
list := sampler sampleNamesSorted.

dsp := sampler stereo asDsp.
dsp init.
dsp start.

dsp setValue: 261 parameter: 'PAMSamplerNote'.

Reciter textToIPAPhonemes: 'book'. "#('b' 'ʊ' 'k')" "#('n' 'i_colon' 'r' '?' 'æ')"
myString := Reciter textToIPAPhonemes: 'book'.

"[myString do: [ : i |  dsp setValue: ( list  indexOf: i)  parameter: 'PAMSamplerIndex' . dsp trig: 'PAMSamplerGate'  . 0.2 seconds wait] ] fork.
"
"6. Play each phoneme and log index"
[
    myString do: [:i | 
        | idx |
        idx := list indexOf: i.
        Transcript
            show: 'Phoneme: ', i;
            show: ' | Index: ', idx printString;
            cr.
        dsp setValue: idx parameter: 'PAMSamplerIndex'.
        dsp trig: 'PAMSamplerGate'.
        (Delay forSeconds: 0.2) wait.
    ].
] fork.

Transcript open.

• as seen --> the index of phoneme is always 0 --> which in our case is a_colon

• ie. the list := sampler sampleNamesSorted. --> and in loop idx := list indexOf: i. --> aint working ( giving index correclty )

• 

• 

the real issue

3 different sortings

sampler := TpSampler new pathToFolder: x ; label: 'PAMSampler'.
sampler sampleNamesSorted.

ipaList := (x asFileReference children collect: [ :f | f basenameWithoutExtension ]).`
ipaListSorted := ipaList asSortedCollection asArray.`

3. method 2 plus adding them to a new folder : prolly OS also sorts them here automatically

ipaList := folder children collect: [ :f | f basenameWithoutExtension ].
ipaListSorted := ipaList asSortedCollection asArray.

"Create a temp folder with files in sorted order" 
sortedFolder := FileLocator documents / 'sorted_phoneme_dataset'. "{temp}\sorted_phoneme_dataset"
sortedFolder ensureCreateDirectory. "{documents}\sorted_phoneme_dataset"

"Copy files in sorted order to temp folder"
ipaListSorted do: [:name |
    | src dest |
    src := folder / (name , '.wav').
    dest := sortedFolder / (name , '.wav').
    src copyTo: dest.
].

ALL THREE METHODS have different results DIFFERENTTTTTTTT !!!!! 😢

Soln 1. : sort using 2 --> find the indexes of phoneme of our IPAString --> provide this list of index to DSP

• issue --> we are still dependent on OS sorting method
  • if for us index 4 == 'w.wav' --> for OS the same index can be 'p.wav' 😦
  • from where is OS sorting file coming in picture ?
    • 
    • Index == order of File structure 😠

| folder ipaList ipaListSorted list dsp myString indexList |

"1. Get folder"
folder := FileLocator documents / 'phoneme_dataset'.

"2. Get sorted list of phoneme names (without extension)"
ipaList := folder children collect: [ :f | f basenameWithoutExtension ].
ipaListSorted := ipaList asSortedCollection asArray.

"Use ipaListSorted as our list"
list := ipaListSorted.

"3. Init DSP from the folder directly"
dsp := (TpSampler new
    pathToFolder: folder fullName;
    label: 'PAMSampler') stereo asDsp.

dsp init.
dsp start.
dsp setValue: 261 parameter: 'PAMSamplerNote'.
dsp traceAllParams.
	
"4. Convert the word to IPA phonemes"
myString := Reciter textToIPAPhonemes: 'book'.  "#('b' 'ʊ' 'k')"

"5. Find the index of each phoneme in the sorted list"
indexList := myString collect: [ :phoneme | list indexOf: phoneme ].

Transcript show: 'Indexes for "book": ', indexList printString; cr.

"6. Feed these indexes to DSP"
[
    indexList do: [:idx |
        Transcript show: 'Playing index ', idx printString; cr.
        dsp setValue: idx parameter: 'PAMSamplerIndex'.
        dsp trig: 'PAMSamplerGate'.
        (Delay forSeconds: 0.3) wait.
    ].
] fork.

Soln 2.

we need to rename these files as number_phoneme -->> to get correct index 💔

• the order in which we wanna name :

• python script that does the job !!! : Rename files as number_phonemeName.wav

  • output of terminal :
    • 
  • Output of the file structure nowwwww !!!!! :
    • 
  • yipeeeee 😎

some other rookie mistakjes in SAM phoenem to IPA phoneme converter :

remove ? from logic of SAM phoenem to IPA phoneme converter

• 

• when we remove stress marker -> we also need to remove \ from the string --> as they can result in ? in IPA Phoneem string.

• 

• code :

• works now :

adding more rules and a '' empty rule in IPA converter. --> to makr the dictionary comprehensive

• hence no more ? in IPA string
  • 

added better loggers tooo !!

on to the actual code !!

• cant use below .. as now we are printing 000_phoneme

  • 

• hence, need to define a dictionary !!

  • 

"the code i want to use"
| path sampler dsp ipaOrder word myString targetIndex phoneme |

"1. Path to your already-sorted folder"
path := FileLocator documents / 'phoneme_dataset_sorted'.

"2. Create sampler from sorted folder"
sampler := TpSampler new
    pathToFolder: path fullName;
    label: 'PAMSampler'.

"3. Define your custom IPA order (matches file order in phoneme_dataset_sorted)"
ipaOrder := #('a_colon' 'aɪ' 'aʊ' 'b' 'd' 'dʒ' 'e' 'eə' 'eɪ' 'f' 'g' 'h' 'i' 'i_colon' 'j' 'k' 'l' 'm' 'n' 'p' 'r' 's' 't' 'u_colon' 'v' 'w' 'z' 'æ' 'ð' 'ŋ' 'ɒ' 'ɔ_colon' 'ɔɪ' 'ə' 'əʊ' 'ɜ_colon' 'ɪə' 'ʃ' 'ʈʃ' 'ʊ' 'ʊə' 'ʌ' 'ʒ' 'θ').

"4. Initialize DSP"
dsp := sampler stereo asDsp.
dsp init.
dsp start.

"Set audible note & level"
dsp setValue: 261 parameter: 'PAMSamplerNote'.   "Middle C"
"dsp setValue: 0.9 parameter: 'PAMSamplerUlevel'. ""Loud enough"""

"5. The word we want to play"
word := 'cat'.
myString := Reciter textToIPAPhonemes: word.

Transcript show: 'Phonemes for "', word, '": ', myString printString; cr.

"6. Play each phoneme in sequence"
[
    myString do: [:phoneme |
        targetIndex := ipaOrder indexOf: phoneme.
        Transcript show: 'Playing ', phoneme, ' (index ', targetIndex printString, ')'; cr.
        dsp setValue: targetIndex - 1 parameter: 'PAMSamplerIndex'.  "0-based index"
        dsp trig: 'PAMSamplerGate' for: 0.05.
        0.25 seconds wait
    ].
] fork.

Transcript open.

refactoring the PAMDSP code completley !!

info that might come handy!

• where should cleanup happen --> in main class / utility class
  • 
• when to stop + destroy dsp in asynchrouns calls
  • 
  • hence, cleanup inside a fork --> so it destroys that threads dsp !!
    • 

Final looks of code now ❤️

| pamDsp dsp|

pamDsp := PAMDsp new. "a PAMDsp"  "This initializes DSP and starts it"
pamDsp sayNumbers0to9 .
pamDsp sayNumbers: #(8 3 5 7).

• AUDIO TIMEEEEE !!!
  • ▶ Watch demo
  • ▶ Watch demo
  • Transcripts :
    • 

11th August'25

some rookie code mistakes + optimization

1. Finding better ways to have specific file paths

    "Get the documents folder and append subfolder in a cross-platform way"
    folderPath := FileLocator documents / aFolderName.

2. why the below error ? --> cos of string concatation

 path := FileLocator documents , pathSeparator , 'numbers'. "File @ C:\Users\Neerja\Documents.\.numbers"

• 
• reolved image :
  • 

3. are we handling indexes correctly via this line ?

ipaText do: [:token |
            idx := sortedIPAarray indexOf: token ifAbsent: [ 0 ].
				PAMDspUtility log: 'Sorted IPA Array index for ', token printString, ' is ', idx printString.

            idx > 0 ifTrue: [
                dsp setValue: idx-1 parameter: 'PhonemeIndex'.
                PAMDspUtility log: ('Playing phoneme ' , token , ' at index ' , idx printString).
                dsp trig: 'PhonemeGate' for: 0.05.
                (Delay forSeconds: 1) wait.
            ] ifFalse: [
                PAMDspUtility log: ('Skipping unknown phoneme token: "' , token , '"').
            ].
        ].

Final SHOWDOWNNNNNNNN

| pamDsp |

pamDsp := PAMDsp new. 
pamDsp sayPhoneme: 'dog' . 

• VIDEOOOO TIMEEE

  • ▶ Watch demo

  • transcript :

    • 

  • NOTE : we later renamed method to pamDsp sayText: 'dog'.

questions for Domenico

• how to set value of Ulevel.
• is there a way i can get the values for of Ulevel, Gate, Index, Note ... for lets say logging
• how to make this better ? --> is it parser ? giving us amplitude + pitch at which we speak

12th August'25

why we cannot hear some audio files in DSP --> but can hear some

• in PAMDsp we have been loading the stereo file only

• apparently they were all mono

• code to find mono or stero : Identify Code

• code to convert mono to stereo : Convert Code

  • terminal output :

• another way to find : RIGHT CLICK ON AUDIO FILE -> PROPERTIES -> CHANNEL

  • 

code to convert mono to stereo :

• need to install ffmeg

  • installaiton link : installation link
  • YT link to install : yt
  • installed succesfully :

• runnign the convert code :

• identify code after convert code :

  • 
  • ITS STEREO NOWWW !! :
    • 

understanding the presets

"value to tweak ~~"

"amplitude ...aka Volume"
dsp setValue: 0.2 parameter: 'PAMSamplerULevel'. "Loud enough"

"pitch ... aka sharpness of voice"
dsp setValue: 500 parameter: 'PAMSamplerNote'.   "Middle C"

"streching the note"
dsp trig: 'PAMSamplerGate' for: 2.

"wait time to play next phoneme"
0.5 seconds wait

understanding params

1. PAMSamplerNote : RANGE : [ 200 to 900 ]

audio smaples : playing with frequency .. aka pitch

• ▶ Watch demo

2. PAMSamplerULevel : RANGE : [ 0 to 1 ]

audio smaples : playing with amplitude .. aka volume

• ▶ Watch demo

3. streching a note

• think of it as hitting a piano key .. and pressing on it
  • if value is 2 , 3.. it eventually fades off
  • if value is 0.1 .. no streching ... best gliding seen between notes !

audio smaples : playing with note streching

• ▶ Watch demo

4. wait time to play next phoneme

• ig 0.4 is good

gender based data

age based voice

understanding the time given in the above code snippet

• 
• how can we strech note ?
  • 

so when we are making our stress + length in parser -- what we mean to apply is :

• stress == volumen added to that phoenme .. ieUlevel ( we can also think of lowering the pitch Note )
• Length == strech of the phoneme .. ie in line dsp trig: 'PAMSamplerGate' for: 0.1.

"the code i want to use"
| path sampler dsp ipaOrder word myString targetIndex phoneme |

"1. Path to your already-sorted folder"
path := FileLocator documents / 'phoneme'.

"2. Create sampler from sorted folder"
sampler := TpSampler new
    pathToFolder: path fullName;
    label: 'PAMSampler'.

"3. Define your custom IPA order (matches file order in phoneme_dataset_sorted)"
ipaOrder := #('a_colon' 'aɪ' 'aʊ' 'b' 'd' 'dʒ' 'e' 'eə' 'eɪ' 'f' 'g' 'h' 'i' 'i_colon' 'j' 'k' 'l' 'm' 'n' 'p' 'r' 's' 't' 'u_colon' 'v' 'w' 'z' 'æ' 'ð' 'ŋ' 'ɒ' 'ɔ_colon' 'ɔɪ' 'ə' 'əʊ' 'ɜ_colon' 'ɪə' 'ʃ' 'ʈʃ' 'ʊ' 'ʊə' 'ʌ' 'ʒ' 'θ').

"4. Initialize DSP"
dsp := sampler stereo asDsp.
dsp init.
dsp start.
dsp displayUI.
dsp traceAllParams.

"Set audible note & level"
dsp setValue: 1 parameter: 'PAMSamplerULevel'. "Loud enough"
dsp setValue: 500 parameter: 'PAMSamplerNote'.   "Middle C"

"5. The word we want to play"
word := 'dog'.
myString := Reciter textToIPAPhonemes: word.

Transcript show: 'Phonemes for "', word, '": ', myString printString; cr.

"6. Play each phoneme in sequence"
[
    myString do: [:phoneme |
        targetIndex := ipaOrder indexOf: phoneme.
        Transcript show: 'Playing ', phoneme, ' (index ', targetIndex printString, ')'; cr.
        dsp setValue: targetIndex - 1 parameter: 'PAMSamplerIndex'.  "0-based index"
        dsp trig: 'PAMSamplerGate' for: 0.1.
        0.4 seconds wait
    ].
] fork.

Transcript open.

• words having g , s , d , u_colon --> are pronoced well [ dog , cap ]

am out till 18th --> see ya then ❤️

18th August'25

finding amplitude and length from stress

• stress value ranges from [ 1 to 5] .. if its 0 --> we are considerign it default 1
• hence 5 === max amplitude , 1 === min amplitude
• in pharo ... i dont want to play it below .5 ... so in pharo sam
  • conversion logic :
  • stress of [1 to 5 ] === sam amplitude [ 0.5 to 1]
  • formula :
    • 

our limits

our formulas !!

• amplitude :

  • 

• duration :

  • 

• output :

  • 

FINALLLLL SHOWWW FOR PROSODDDYYYYYY !!!!! ❤️

|dsp|

dsp := PAMDsp new.

"dsp sayText: 'dog'."

dsp sayTextWithProsody: 'dog'.

• ▶ Watch demo
• commit :
• Trascript : 😇
  • 
  • yellow box :[ ipa_phoneem , stress value ]
  • green box : [ amplitude , stress , duration parameters ]

doing age based initlizataion !!!

FINALLLLL PROSODYYYYYY SHOW BASED ON AGEEEE !!!!!! 😗

|dsp|

dsp := PAMDsp new.

"dsp sayText: 'dog'."

"toggle true / false below"
dsp sayTextWithProsody: 'dog' asChild: false.

• ▶ Watch demo
• how we did it ?

  • pass as a flag in function :

    • 

  • the helper utility used above:

    • 

  • in PAMDSPUtility we defined instanceVariables :

    • 

    • and respective getter-setter

      • 
      • similar with adultPreset

    • so the note value returns from PAMDSPUtility and is set as our Ulevel in our PAMDSP !!!! 😍

    • Transcript :

      • 

shagun ke 101 commitssss !!!!! -- happpyyyy dayyyyy 😂

23rd August'25

• quick meet with Domenico

expectationos in last week of GSoC ie. 25th to 31st Aug

• need to put the folder of [ phoneme + number ] in GitHub

• write a little readme

  • history of PAM
  • link SAM javascript .. inspired by this
  • made via phausto, using this samples ---> phoneme + numbers
  • examples code snippets

• ToDo :

  • code changes
    • long method warning in code --> make short methods
    • put comments in class
    • use streams instead of strings : refer pharo books

things to write in Submission :

• in future dev of project : we can have formant synthesis based TTS

issues resolved

pragma

• for pragma .. think we are writing in a playground but without self
• the escape character for string ie. '' is ''
• hence write the praga defnition as :

<script: 'self sayTextWithProsody: ''dog'' asChild: false'> : escape charater

baseline

• each package we are adding ie [Coypu, CoypuIDE, Phasuto] is calling different version of Toplo

• just add dependency of CoypuIDE

  • Phasto need CoypuIDE
  • we need Phausto
  • so we add only phausto

• 

• chekc wiki of phausto :

• 

resolving baseline issue post meet

• adding just CoypuIDE in baseline stil failed 😢
  

• changed our metacello script to :

Metacello new
	baseline: 'PAM';
  repository: 'github://neerja-1984/PAM-GSoC-25-Project:master/src';
  onConflict: [ :ex | ex useIncoming ];
 	onUpgrade: [ :ex | ex useIncoming ];
  load.

• use the incoming version when conflict arises, then upgarde to in !!!

It worked in a new imageeee + all tests are greenn !!!! 💚💚💚💚💚💚💚

• ▶ Watch demo
• 

24th August'25

Blog timeeeee !!

• ideas as of now in my mind`

1. in PAM code project

• add some code snippets + basic info + add zips of audio folder to it
• make a wiki for it for metacello installation

2. make another repo : Relive Gsoc'25 with me

• add readme of weekly updates
• make a github.io ( website of it ) --> add gsoc submission related data + blog links
• see past submissions of previous gsoc students

refernce :

• https://github.com/valens200/GSoC-2024
• https://tyler-li-mcgill.tech/blog/gsoc3

26th August'25

Gthub Hosted page !!! 😊

• Click Here to watch it live !!
• Github Repo

uploaded this awesomeeee readme to Github with images and videos

• PS. Videos dont load -> coz of high memory ( viewer will have to download it to view from Github ) 💔

want to make a lovely PDF of this lovely readme i prepared in the last few months!! ❤️😭

• vscode extension -> Markdown PDF (https://marketplace.visualstudio.com/items?itemName=yzane.markdown-pdf)
• right click readme -> export to pdf
• creates a readme.pdf next to readme.md in folder structure 😄

27th August'25

counting number of lines on a github repo

• first clone a repo somewher :
  • git clone https://github.com/neerja-1984/PAM-GSoC-25-Project.git
  • run in gitbash : git ls-files | xargs wc -l
  • 
  • total lines : 5053

to add in Blog :

• numbers :
  • ▶ Watch demo
  • transcript :
• text :
  • ▶ Watch demo
  • transcript :

30th August'25

added audio samples zip to the github.io [rewind-gsoc25] page + added as a release to the github repo [PAM-GSoC-25-Project]

• rewind-gsoc25 repo
• PAM-GSoC-25-Project

Finallyyyy done with the github.io page ! Lookss fabulous ❤️

• Link : https://neerja-1984.github.io/rewind-gsoc25/

added readme section in PAM-GSoC-25-Project

6th Sept'25

final meet ! 😢

happy pics ... smiley faces !

Domenico said --> they misjudged the project timeline --> project size should've been more + this is the max that could have been done in 3 months time

showed them github.io page + live demo ... they were happyy and satisfied !

Nahuel was happy that project has 110 commits !!

next year .. improving PAM project

Nahuel pointed out --> remember the first time u were struglging with Pharo .. and see the project u have built from scratch

everythignmakes sense when we look back and see everything we have done ❤️

really happy day !

IMPT !!!

To make push in the github repo [rewind-gsoc25]

• add files from vscode extension
• in terminal go cd C:\practicals\code\GSOC Proposal\Github-pages-website\assets
• git commit -m "message"
• git push

29th September'25

• added 2nd blog ---> oh gawdd the procrastination :(

• my favourite links !!! ❤️

  • GSOC portfolio link
  • Rewind Gsoc 25 link

• both blogs :

  • the hows of written english to spoken english
  • lets make a lightweight text to speech model because why not