phoonnx 0.0.2a1__py3-none-any.whl → 0.1.0a1__py3-none-any.whl

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/phonemizers/ar.py

@@ -1,11 +1,15 @@
 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):
-        super().__init__(Alphabet.IPA)
+    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:
@@ -26,17 +30,77 @@ class MantoqPhonemizer(BasePhonemizer):
 
     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)
-        return "".join(phonemes).replace("_+_", " ")
+
+        # 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__":
-    text = "مرحبا بالعالم"
-    # gets normalized to
-    # مَرْحَبًا بِالْعالَم
+    # 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)
 
-    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)]
+    # 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/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/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/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.dist-info → phoonnx-0.1.0a1.dist-info}/METADATA

@@ -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
@@ -8,7 +8,7 @@ Requires-Dist: numpy
 Requires-Dist: onnxruntime
 Requires-Dist: quebra-frases
 Requires-Dist: langcodes
-Requires-Dist: ovos-number-parser>=0.3.3a1
+Requires-Dist: ovos-number-parser>=0.4.0
 Requires-Dist: ovos-date-parser>=0.6.4a1
 Provides-Extra: aa
 Requires-Dist: epitran; extra == "aa"
@@ -214,6 +214,12 @@ Provides-Extra: tpi
 Requires-Dist: epitran; extra == "tpi"
 Provides-Extra: tr
 Requires-Dist: epitran; extra == "tr"
+Provides-Extra: train
+Requires-Dist: cython<1,>=0.29.0; extra == "train"
+Requires-Dist: librosa<1,>=0.9.2; extra == "train"
+Requires-Dist: numpy<2,>=1.19.0; extra == "train"
+Requires-Dist: pytorch-lightning<2.0; extra == "train"
+Requires-Dist: torch<2,>=1.11.0; extra == "train"
 Provides-Extra: uew
 Requires-Dist: epitran; extra == "uew"
 Provides-Extra: ug

{phoonnx-0.0.2a1.dist-info → phoonnx-0.1.0a1.dist-info}/RECORD

@@ -1,19 +1,19 @@
 phoonnx/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-phoonnx/config.py,sha256=bO7dx2tfLotkohict3UKlCEVm-BRFB1feYYR1HarUkk,19382
+phoonnx/config.py,sha256=81H34oPG2BaiOA6UM1KapoT341n068LqRprKb5ER6mY,19451
 phoonnx/phoneme_ids.py,sha256=FiNgZwV6naEsBh6XwFLh3_FyOgPiCsK9qo7S0v-CmI4,13667
 phoonnx/util.py,sha256=XSjFEoqSFcujFTHxednacgC9GrSYyF-Il5L6Utmxmu4,25909
-phoonnx/version.py,sha256=tnwL6H7F2rKiYfzZqG80Wt82MT0mHbzZ38RjwaJcsgs,114
+phoonnx/version.py,sha256=95gLFCt-8xv9DgF7FIF6CljWmhm8SUhevumEBfo7Pl0,114
 phoonnx/voice.py,sha256=FR_LafK1vSi_anPERJjZBuH3Bb9vUIof0MAW6TnALlA,20024
 phoonnx/locale/ca/phonetic_spellings.txt,sha256=igv3t7jxLSRE5GHsdn57HOpxiWNcEmECPql6m02wbO0,47
 phoonnx/locale/en/phonetic_spellings.txt,sha256=xGQlWOABLzbttpQvopl9CU-NnwEJRqKx8iuylsdUoQA,27
 phoonnx/locale/gl/phonetic_spellings.txt,sha256=igv3t7jxLSRE5GHsdn57HOpxiWNcEmECPql6m02wbO0,47
 phoonnx/locale/pt/phonetic_spellings.txt,sha256=KntS8QMynEJ5A3Clvcjq4qlmL-ThSbhfD6v0nKSrlqs,49
 phoonnx/phonemizers/__init__.py,sha256=QGBZk0QUgJdg2MwUWY9Kpk6ucwrEJYtHb07YcNvXCV4,1647
-phoonnx/phonemizers/ar.py,sha256=W3wJ_GI8gV3qb9fng5xeazX8ZlY-aNSywATTdxryiZg,1257
+phoonnx/phonemizers/ar.py,sha256=29bCfYhlhx0QX3PQyx3EkUghzh8YfkxNAnMAICXX6I8,4148
 phoonnx/phonemizers/base.py,sha256=yPg6-dvscYpl3rR3JEULG1PRF-i8DWC_C3HAZGLbxOo,7648
 phoonnx/phonemizers/en.py,sha256=N2SVoVhplQao7Ej5TXbxJU-YkAgkY0Fr9iYBFnsjFSE,9271
 phoonnx/phonemizers/fa.py,sha256=d_DZM2wqomf4gcRH_rFcNA3VkQWKHru8vwBwaNG8Ll8,1452
-phoonnx/phonemizers/gl.py,sha256=05DX_9lQa3Uws3RPcWPtw0nXX6Vo3bybrZkpMErguTo,5529
+phoonnx/phonemizers/gl.py,sha256=jEFKJJViHufZtB7lGNwWQCdWGiNKDCVZ_GRYXTaw_2c,6614
 phoonnx/phonemizers/he.py,sha256=KbRI3XRZa8UtJdNWmn_fd-t5lmFSIp4Mw8UgcO5l-Po,2211
 phoonnx/phonemizers/ja.py,sha256=Xojsrt715ihnIiEk9K6giYqDo9Iykw-SHfIidrHtHSU,3834
 phoonnx/phonemizers/ko.py,sha256=kwWoOFqanCB8kv2JRx17A0hP78P1wbXlX6e8VBn1ezQ,2989
@@ -22,6 +22,7 @@ phoonnx/phonemizers/vi.py,sha256=_XJc-Xeawr1Lxr7o8mE_hJao1aGcj4g01XYAOxC_Scg,131
 phoonnx/phonemizers/zh.py,sha256=88Ywq8h9LDanlyz8RHjRSCY_PRK_Dq808tBADyrgaP8,9657
 phoonnx/thirdparty/__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
 phoonnx/thirdparty/arpa2ipa.py,sha256=Uj1G5NgP5oBBfSm26LGB8QoumdT-NqCLQTZHT165-_o,5850
+phoonnx/thirdparty/bw2ipa.py,sha256=5FiWC4AP4KXkqtbclbinoXEsUnSYEjk4VWAPasMMcbg,2328
 phoonnx/thirdparty/hangul2ipa.py,sha256=e2c0WOy5lFMcf6GS7pNqIbauMKBX07S84lCczZAZJGA,27518
 phoonnx/thirdparty/zh_num.py,sha256=SESA6gvSJW3LZ0FLoybXn2SpbxqhQTi9Tg_U2IZ5JYY,7147
 phoonnx/thirdparty/cotovia/cotovia_aarch64,sha256=BsAWZN452Lm9kDU4i6rQGHFSlmxP3GfHRKhbJMUQrfA,6764592
@@ -70,7 +71,7 @@ phoonnx_train/vits/attentions.py,sha256=yc_ViF8zR8z68DzphmVVVn27f9xK_5wi8S4ITLXV
 phoonnx_train/vits/commons.py,sha256=JsD8CdZ3ZcYYubYhw8So5hICBziFlCrKLrv1lMDRCDM,4645
 phoonnx_train/vits/config.py,sha256=oSuUIhw9Am7BQ5JwDgtCO-P1zRyN7nPgR-U1XuncJls,10789
 phoonnx_train/vits/dataset.py,sha256=DLLGSCkn3GF9uktoTprH1ERblZ18GO6-QsClQKWa98o,6804
-phoonnx_train/vits/lightning.py,sha256=iJJNAT1-jm0wYvTfBy9T_eZapDN62Pny9YdHEpnApso,12298
+phoonnx_train/vits/lightning.py,sha256=ZBuSIiJ7EUU1Za2V8Uh6-_HGGRW_qwpXLLs1cEDirHA,12301
 phoonnx_train/vits/losses.py,sha256=j-uINhBcYxVXFvFutiewQpTuw-qF-J6M6hdJVeOKqNE,1401
 phoonnx_train/vits/mel_processing.py,sha256=huIjbQgewSmM39hdzRZvZUCI7fTNSMmLcAv3f8zYb8k,3956
 phoonnx_train/vits/models.py,sha256=9PziprRtkdBQ6AowFe1vG4QTCk02By-LDS9W8EtZGvE,24303
@@ -80,7 +81,7 @@ phoonnx_train/vits/utils.py,sha256=exiyrtPHbnnGvcHWSbaH9-gR6srH5ZPHlKiqV2IHUrQ,4
 phoonnx_train/vits/wavfile.py,sha256=oQZiTIrdw0oLTbcVwKfGXye1WtKte6qK_52qVwiMvfc,26396
 phoonnx_train/vits/monotonic_align/__init__.py,sha256=5IdAOD1Z7UloMb6d_9NRFsXoNIjEQ3h9mvOSh_AtO3k,636
 phoonnx_train/vits/monotonic_align/setup.py,sha256=0K5iJJ2mKIklx6ncEfCQS34skm5hHPiz9vRlQEvevvY,266
-phoonnx-0.0.2a1.dist-info/METADATA,sha256=i-4sLAwReU6JT44xM2JVBcAfKP0Rv9mMpWpiEdNUX7U,7868
-phoonnx-0.0.2a1.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
-phoonnx-0.0.2a1.dist-info/top_level.txt,sha256=ZrnHXe-4HqbOSX6fbdY-JiP7YEu2Bok9T0ji351MrmM,22
-phoonnx-0.0.2a1.dist-info/RECORD,,
+phoonnx-0.1.0a1.dist-info/METADATA,sha256=9FZiRhA48da6ZbX1qCrKKVqsWMWQwfedz-bUXATd6Sk,8145
+phoonnx-0.1.0a1.dist-info/WHEEL,sha256=tZoeGjtWxWRfdplE7E3d45VPlLNQnvbKiYnx7gwAy8A,92
+phoonnx-0.1.0a1.dist-info/top_level.txt,sha256=ZrnHXe-4HqbOSX6fbdY-JiP7YEu2Bok9T0ji351MrmM,22
+phoonnx-0.1.0a1.dist-info/RECORD,,

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