phoonnx 0.0.2a1__tar.gz → 0.1.0a1__tar.gz

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: phoonnx
-Version: 0.0.2a1
+Version: 0.1.0a1
 Home-page: https://github.com/TigreGotico/phoonnx
 Author: JarbasAi
 Author-email: jarbasai@mailfence.com
@@ -71,6 +71,7 @@ Provides-Extra: got
 Provides-Extra: tpi
 Provides-Extra: pt
 Provides-Extra: ckb
+Provides-Extra: train
 Provides-Extra: lsm
 Provides-Extra: fa
 Provides-Extra: av

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx/config.py

@@ -6,7 +6,6 @@ from phoonnx.phoneme_ids import (load_phoneme_ids, BlankBetween,
                                  DEFAULT_BLANK_WORD_TOKEN, DEFAULT_BLANK_TOKEN,
                                  DEFAULT_PAD_TOKEN, DEFAULT_BOS_TOKEN, DEFAULT_EOS_TOKEN)
 
-
 DEFAULT_NOISE_SCALE = 0.667
 DEFAULT_LENGTH_SCALE = 1.0
 DEFAULT_NOISE_W_SCALE = 0.8
@@ -22,6 +21,8 @@ class Alphabet(str, Enum):
     UNICODE = "unicode"
     IPA = "ipa"
     ARPA = "arpa" # en
+    SAMPA = "sampa"
+    XSAMPA = "x-sampa"
     HANGUL = "hangul" # ko
     KANA = "kana" # ja
     HIRA = "hira" # ja
@@ -32,6 +33,7 @@ class Alphabet(str, Enum):
     ERAAB = "eraab" # fa
     COTOVIA = "cotovia" # gl
     HANZI = "hanzi" # zh
+    MANTOQ = "mantoq" # ar
 
 
 

phoonnx-0.1.0a1/phoonnx/phonemizers/ar.py

@@ -0,0 +1,106 @@
+from phoonnx.phonemizers.base import BasePhonemizer
+from phoonnx.thirdparty.mantoq import g2p as mantoq
+from phoonnx.config import Alphabet
+from phoonnx.thirdparty.bw2ipa import translate as bw2ipa
+
+
+class MantoqPhonemizer(BasePhonemizer):
+
+    def __init__(self, alphabet=Alphabet.MANTOQ):
+        if alphabet not in [Alphabet.IPA, Alphabet.MANTOQ]:
+            raise ValueError("unsupported alphabet")
+        super().__init__(alphabet)
+
+    @classmethod
+    def get_lang(cls, target_lang: str) -> str:
+        """
+        Validates and returns the closest supported language code.
+
+        Args:
+            target_lang (str): The language code to validate.
+
+        Returns:
+            str: The validated language code.
+
+        Raises:
+            ValueError: If the language code is unsupported.
+        """
+        # this check is here only to throw an exception if invalid language is provided
+        return cls.match_lang(target_lang, ["ar"])
+
+    def phonemize_string(self, text: str, lang: str = "ar") -> str:
+        """
+        Phonemizes an Arabic string using the Mantoq G2P tool.
+        If the alphabet is set to IPA, it then converts the result using bw2ipa.
+        """
+        lang = self.get_lang(lang)
+        # The mantoq function returns a tuple of (normalized_text, phonemes)
+        normalized_text, phonemes = mantoq(text)
+
+        # The phonemes are a list of characters, we join them into a string
+        # and replace the word separator token with a space.
+        phonemes =  "".join(phonemes).replace("_+_", " ")
+
+        if self.alphabet == Alphabet.IPA:
+            # If the alphabet is IPA, we use the bw2ipa function to translate
+            # the Buckwalter-like phonemes into IPA.
+            return bw2ipa(phonemes)
+
+        # Otherwise, we return the phonemes in the default Mantoq alphabet.
+        return phonemes
+
+
+if __name__ == "__main__":
+    # Initialize phonemizers for both MANTOQ and IPA alphabets
+    pho_mantoq = MantoqPhonemizer()
+    pho_ipa = MantoqPhonemizer(alphabet=Alphabet.IPA)
+
+    text1 = "مرحبا بالعالم"
+    print(f"Original Text: {text1}")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text1, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text1, 'ar')}")
+    print("-" * 20)
+
+    text2 = "ذهب الطالب إلى المكتبة لقراءة كتاب عن تاريخ الأندلس."
+    print(f"Original Text: {text2}")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text2, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text2, 'ar')}")
+    print("-" * 20)
+
+    # --- New Test Cases to check bw2ipa logic ---
+    print("--- New Test Cases for bw2ipa logic ---")
+
+    # 1. Test for gemination of a sun letter (e.g., ash-shams)
+    text3 = "الشمس"
+    print(f"Original Text: '{text3}'")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text3, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text3, 'ar')}")
+    print("-" * 20)
+
+    # 2. Test for long vowels (e.g., 'fil' - elephant)
+    text4 = "فيل"
+    print(f"Original Text: '{text4}'")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text4, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text4, 'ar')}")
+    print("-" * 20)
+
+    # 3. Test for glide (e.g., 'yawm' - day)
+    text5 = "يوم"
+    print(f"Original Text: '{text5}'")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text5, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text5, 'ar')}")
+    print("-" * 20)
+
+    # 4. Test for long vowels (e.g., 'suwr' - wall)
+    text6 = "سور"
+    print(f"Original Text: '{text6}'")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text6, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text6, 'ar')}")
+    print("-" * 20)
+
+    # 5. Test for glide (e.g., 'law' - if)
+    text7 = "لو"
+    print(f"Original Text: '{text7}'")
+    print(f"  Mantoq Phonemizer: {pho_mantoq.phonemize_string(text7, 'ar')}")
+    print(f"  IPA Phonemizer:    {pho_ipa.phonemize_string(text7, 'ar')}")
+    print("-" * 20)

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx/phonemizers/gl.py

@@ -12,6 +12,57 @@ class CotoviaError(Exception):
     pass
 
 
+COTOVIA2IPA = {
+    "pau": " ",
+    "a": "a",
+    "E": "ɛ",
+    "e": "e",
+    "i": "i",
+    "j": "j",
+    "O": "ɔ",
+    "o": "o",
+    "u": "u",
+    "w": "w",
+    "p": "p",
+    "b": "b",
+    "B": "β",
+    "t": "t",
+    "d": "d",
+    "D": "ð",
+    "k": "k",
+    "g": "g",
+    "G": "ɣ",
+    "f": "f",
+    "T": "θ",
+    "s": "s",
+    "S": "ʃ",
+    "tS": "tʃ",
+    "m": "m",
+    "n": "n",
+    "J": "ɲ",
+    "N": "ŋ",
+    "l": "l",
+    "Z": "ʎ",
+    "jj": "ʎ",
+    "L": "ʎ",
+    "r": "ɾ",
+    "rr": "r",
+    "X": "x"
+}
+
+
+def cotovia2ipa(text: str) -> str:
+    """
+    Converts a string of Cotovía phonemes to IPA.
+    """
+    # Sort the dictionary keys by length in descending order to handle multi-character phonemes first
+    sorted_cotovia_keys = sorted(COTOVIA2IPA.keys(), key=len, reverse=True)
+    ipa_str = text
+    for cotovia_char in sorted_cotovia_keys:
+        ipa_str = ipa_str.replace(cotovia_char, COTOVIA2IPA[cotovia_char])
+    return ipa_str
+
+
 class CotoviaPhonemizer(BasePhonemizer):
     """
     A phonemizer class that uses the Cotovia TTS binary to convert text into phonemes.
@@ -19,7 +70,7 @@ class CotoviaPhonemizer(BasePhonemizer):
     regular expression transformations to clean and normalize the phonetic representation.
     """
 
-    def __init__(self, cotovia_bin_path: Optional[str] = None):
+    def __init__(self, cotovia_bin_path: Optional[str] = None, alphabet: Alphabet = Alphabet.IPA):
         """
         Initializes the CotoviaPhonemizer.
 
@@ -31,7 +82,7 @@ class CotoviaPhonemizer(BasePhonemizer):
         if not os.path.exists(self.cotovia_bin):
             raise FileNotFoundError(f"Cotovia binary not found at {self.cotovia_bin}. "
                                     "Please ensure it's installed or provide the correct path.")
-        super().__init__(Alphabet.COTOVIA)
+        super().__init__(alphabet)
 
     @classmethod
     def get_lang(cls, target_lang: str) -> str:
@@ -127,6 +178,8 @@ class CotoviaPhonemizer(BasePhonemizer):
         # substitute ' ( text )' to ', text,'
         str_ext = re.sub(r"(\w+)\s*\(\s*([^\(\)]*?)\s*\)", r"\1, \\2,", str_ext)
 
+        if self.alphabet == Alphabet.IPA:
+            return cotovia2ipa(str_ext)
         return str_ext
 
 
@@ -138,5 +191,5 @@ if __name__ == "__main__":
     lang = "gl"
     text_gl = "Este é un sistema de conversión de texto a voz en lingua galega baseado en redes neuronais artificiais. Ten en conta que as funcionalidades incluídas nesta páxina ofrécense unicamente con fins de demostración. Se tes algún comentario, suxestión ou detectas algún problema durante a demostración, ponte en contacto connosco."
     print(f"\n--- Getting phonemes for '{text_gl}' (Cotovia) ---")
-    phonemes_cotovia = cotovia.phonemize(text_gl, lang)
+    phonemes_cotovia = cotovia.phonemize_string(text_gl, lang)
     print(f"  Cotovia Phonemes: {phonemes_cotovia}")

phoonnx-0.1.0a1/phoonnx/thirdparty/bw2ipa.py

@@ -0,0 +1,66 @@
+# -*- coding: UTF8 -*-
+"""
+This script translates Buckwalter-transcribed Modern Standard Arabic to IPA.
+
+This relies on Mantoq tokenization that uses separate tokens for vowel length and consonant gemination.
+
+by Casimiro Ferreira with help of Gemini July 2025.
+"""
+
+import re
+
+# This dictionary maps a single Buckwalter character to its most common IPA equivalent.
+char_dict = {
+    'a': 'a', 'A': 'aː', 'b': 'b', 'c': 'x', 'd': 'd', 'D': 'dˤ', 'e': 'e', 'E': 'ʕ',
+    'f': 'f', 'g': 'ɣ', 'h': 'h', 'H': 'ħ', 'i': 'i', 'I': 'iː', 'j': 'ʒ', 'k': 'k',
+    'l': 'l', 'm': 'm', 'n': 'n', 'p': 'p', 'q': 'q', 'r': 'r', 'R': 'r', 's': 's',
+    'S': 'sˤ', 't': 't', 'T': 'tˤ', 'u': 'u', 'U': 'uː', 'v': 'v', 'w': 'w', 'x': 'x',
+    'y': 'j', 'z': 'z', 'Z': 'ðˤ', '\'': 'ʔ', '<': 'ʔ', 'o': 'o', '-': ' ',
+    '*': 'ð', '$': 'ʃ'
+}
+_vowels = {'a', 'i', 'u', 'aː', 'iː', 'uː'}
+
+
+def translate(buckwalter_text: str) -> str:
+    """
+    Translates a Buckwalter-transcribed string into an IPA string.
+
+    Args:
+        buckwalter_text (str): The Buckwalter string to translate.
+
+    Returns:
+        str: The translated IPA string.
+    """
+    ipa_list = []
+    i = 0
+    while i < len(buckwalter_text):
+        # Check for the longest token first. The new Mantoq tokenization
+        # seems to use a 5-character token.
+        token = buckwalter_text[i:i + 5]
+
+        # If the previous character was a vowel, we assume it's a long vowel
+        # marker (ː). Otherwise, we assume it's a geminated consonant.
+        if token == '_dbl_':
+            if ipa_list and ipa_list[-1] in _vowels:
+                ipa_list.append('ː')  # Add length marker for long vowels
+            elif ipa_list:
+                ipa_list.append(ipa_list[-1])  # Duplicate the consonant
+            i += 5
+            continue
+
+        # Check for multi-character mappings from char_dict
+        two_char_token = buckwalter_text[i:i + 2]
+        if two_char_token in char_dict:
+            ipa_list.append(char_dict[two_char_token])
+            i += 2
+            continue
+
+        # Handle single characters
+        single_char = buckwalter_text[i]
+        if single_char in char_dict:
+            ipa_list.append(char_dict[single_char])
+        else:
+            ipa_list.append(single_char)
+        i += 1
+
+    return ''.join(ipa_list)

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx/version.py

@@ -1,6 +1,6 @@
 # START_VERSION_BLOCK
 VERSION_MAJOR = 0
-VERSION_MINOR = 0
-VERSION_BUILD = 2
+VERSION_MINOR = 1
+VERSION_BUILD = 0
 VERSION_ALPHA = 1
 # END_VERSION_BLOCK

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx.egg-info/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: phoonnx
-Version: 0.0.2a1
+Version: 0.1.0a1
 Home-page: https://github.com/TigreGotico/phoonnx
 Author: JarbasAi
 Author-email: jarbasai@mailfence.com
@@ -71,6 +71,7 @@ Provides-Extra: got
 Provides-Extra: tpi
 Provides-Extra: pt
 Provides-Extra: ckb
+Provides-Extra: train
 Provides-Extra: lsm
 Provides-Extra: fa
 Provides-Extra: av

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx.egg-info/SOURCES.txt

@@ -29,6 +29,7 @@ phoonnx/phonemizers/vi.py
 phoonnx/phonemizers/zh.py
 phoonnx/thirdparty/__init__.py
 phoonnx/thirdparty/arpa2ipa.py
+phoonnx/thirdparty/bw2ipa.py
 phoonnx/thirdparty/hangul2ipa.py
 phoonnx/thirdparty/zh_num.py
 phoonnx/thirdparty/cotovia/cotovia_aarch64
@@ -93,6 +94,7 @@ phoonnx_train/../phoonnx/phonemizers/vi.py
 phoonnx_train/../phoonnx/phonemizers/zh.py
 phoonnx_train/../phoonnx/thirdparty/__init__.py
 phoonnx_train/../phoonnx/thirdparty/arpa2ipa.py
+phoonnx_train/../phoonnx/thirdparty/bw2ipa.py
 phoonnx_train/../phoonnx/thirdparty/hangul2ipa.py
 phoonnx_train/../phoonnx/thirdparty/zh_num.py
 phoonnx_train/../phoonnx/thirdparty/cotovia/cotovia_aarch64

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx.egg-info/requires.txt

@@ -2,7 +2,7 @@ numpy
 onnxruntime
 quebra-frases
 langcodes
-ovos-number-parser>=0.3.3a1
+ovos-number-parser>=0.4.0
 ovos-date-parser>=0.6.4a1
 
 [aa]
@@ -300,6 +300,13 @@ epitran
 [tr]
 epitran
 
+[train]
+cython<1,>=0.29.0
+librosa<1,>=0.9.2
+numpy<2,>=1.19.0
+pytorch-lightning<2.0
+torch<2,>=1.11.0
+
 [uew]
 epitran
 

{phoonnx-0.0.2a1 → phoonnx-0.1.0a1}/phoonnx_train/vits/lightning.py

@@ -299,9 +299,9 @@ class VitsModel(pl.LightningModule):
             test_audio = test_audio * (1.0 / max(0.01, abs(test_audio.max())))
 
             tag = test_utt.text or str(utt_idx)
-            self.logger.experiment.add_audio(
-                tag, test_audio, sample_rate=self.hparams.sample_rate
-            )
+           # self.logger.experiment.add_audio(
+           #     tag, test_audio, sample_rate=self.hparams.sample_rate
+           # )
 
         return val_loss
 

phoonnx-0.0.2a1/phoonnx/phonemizers/ar.py

@@ -1,42 +0,0 @@
-from phoonnx.phonemizers.base import BasePhonemizer
-from phoonnx.thirdparty.mantoq import g2p as mantoq
-from phoonnx.config import Alphabet
-
-class MantoqPhonemizer(BasePhonemizer):
-
-    def __init__(self):
-        super().__init__(Alphabet.IPA)
-
-    @classmethod
-    def get_lang(cls, target_lang: str) -> str:
-        """
-        Validates and returns the closest supported language code.
-
-        Args:
-            target_lang (str): The language code to validate.
-
-        Returns:
-            str: The validated language code.
-
-        Raises:
-            ValueError: If the language code is unsupported.
-        """
-        # this check is here only to throw an exception if invalid language is provided
-        return cls.match_lang(target_lang, ["ar"])
-
-    def phonemize_string(self, text: str, lang: str = "ar") -> str:
-        """
-        """
-        lang = self.get_lang(lang)
-        normalized_text, phonemes = mantoq(text)
-        return "".join(phonemes).replace("_+_", " ")
-
-
-if __name__ == "__main__":
-    text = "مرحبا بالعالم"
-    # gets normalized to
-    # مَرْحَبًا بِالْعالَم
-
-    pho = MantoqPhonemizer()
-    print(pho.phonemize(text, "ar"))
-    # [('m a r H a b a n aa   b i l E aa l a m', '.', True)]