xinference 0.16.3__py3-none-any.whl → 1.2.1__py3-none-any.whl

xinference/thirdparty/melo/text/english.py

@@ -0,0 +1,284 @@
+import pickle
+import os
+import re
+from g2p_en import G2p
+
+from . import symbols
+
+from .english_utils.abbreviations import expand_abbreviations
+from .english_utils.time_norm import expand_time_english
+from .english_utils.number_norm import normalize_numbers
+from .japanese import distribute_phone
+
+from transformers import AutoTokenizer
+
+current_file_path = os.path.dirname(__file__)
+CMU_DICT_PATH = os.path.join(current_file_path, "cmudict.rep")
+CACHE_PATH = os.path.join(current_file_path, "cmudict_cache.pickle")
+_g2p = G2p()
+
+arpa = {
+    "AH0",
+    "S",
+    "AH1",
+    "EY2",
+    "AE2",
+    "EH0",
+    "OW2",
+    "UH0",
+    "NG",
+    "B",
+    "G",
+    "AY0",
+    "M",
+    "AA0",
+    "F",
+    "AO0",
+    "ER2",
+    "UH1",
+    "IY1",
+    "AH2",
+    "DH",
+    "IY0",
+    "EY1",
+    "IH0",
+    "K",
+    "N",
+    "W",
+    "IY2",
+    "T",
+    "AA1",
+    "ER1",
+    "EH2",
+    "OY0",
+    "UH2",
+    "UW1",
+    "Z",
+    "AW2",
+    "AW1",
+    "V",
+    "UW2",
+    "AA2",
+    "ER",
+    "AW0",
+    "UW0",
+    "R",
+    "OW1",
+    "EH1",
+    "ZH",
+    "AE0",
+    "IH2",
+    "IH",
+    "Y",
+    "JH",
+    "P",
+    "AY1",
+    "EY0",
+    "OY2",
+    "TH",
+    "HH",
+    "D",
+    "ER0",
+    "CH",
+    "AO1",
+    "AE1",
+    "AO2",
+    "OY1",
+    "AY2",
+    "IH1",
+    "OW0",
+    "L",
+    "SH",
+}
+
+
+def post_replace_ph(ph):
+    rep_map = {
+        "：": ",",
+        "；": ",",
+        "，": ",",
+        "。": ".",
+        "！": "!",
+        "？": "?",
+        "\n": ".",
+        "·": ",",
+        "、": ",",
+        "...": "…",
+        "v": "V",
+    }
+    if ph in rep_map.keys():
+        ph = rep_map[ph]
+    if ph in symbols:
+        return ph
+    if ph not in symbols:
+        ph = "UNK"
+    return ph
+
+
+def read_dict():
+    g2p_dict = {}
+    start_line = 49
+    with open(CMU_DICT_PATH) as f:
+        line = f.readline()
+        line_index = 1
+        while line:
+            if line_index >= start_line:
+                line = line.strip()
+                word_split = line.split("  ")
+                word = word_split[0]
+
+                syllable_split = word_split[1].split(" - ")
+                g2p_dict[word] = []
+                for syllable in syllable_split:
+                    phone_split = syllable.split(" ")
+                    g2p_dict[word].append(phone_split)
+
+            line_index = line_index + 1
+            line = f.readline()
+
+    return g2p_dict
+
+
+def cache_dict(g2p_dict, file_path):
+    with open(file_path, "wb") as pickle_file:
+        pickle.dump(g2p_dict, pickle_file)
+
+
+def get_dict():
+    if os.path.exists(CACHE_PATH):
+        with open(CACHE_PATH, "rb") as pickle_file:
+            g2p_dict = pickle.load(pickle_file)
+    else:
+        g2p_dict = read_dict()
+        cache_dict(g2p_dict, CACHE_PATH)
+
+    return g2p_dict
+
+
+eng_dict = get_dict()
+
+
+def refine_ph(phn):
+    tone = 0
+    if re.search(r"\d$", phn):
+        tone = int(phn[-1]) + 1
+        phn = phn[:-1]
+    return phn.lower(), tone
+
+
+def refine_syllables(syllables):
+    tones = []
+    phonemes = []
+    for phn_list in syllables:
+        for i in range(len(phn_list)):
+            phn = phn_list[i]
+            phn, tone = refine_ph(phn)
+            phonemes.append(phn)
+            tones.append(tone)
+    return phonemes, tones
+
+
+def text_normalize(text):
+    text = text.lower()
+    text = expand_time_english(text)
+    text = normalize_numbers(text)
+    text = expand_abbreviations(text)
+    return text
+
+model_id = 'bert-base-uncased'
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+def g2p_old(text):
+    tokenized = tokenizer.tokenize(text)
+    # import pdb; pdb.set_trace()
+    phones = []
+    tones = []
+    words = re.split(r"([,;.\-\?\!\s+])", text)
+    for w in words:
+        if w.upper() in eng_dict:
+            phns, tns = refine_syllables(eng_dict[w.upper()])
+            phones += phns
+            tones += tns
+        else:
+            phone_list = list(filter(lambda p: p != " ", _g2p(w)))
+            for ph in phone_list:
+                if ph in arpa:
+                    ph, tn = refine_ph(ph)
+                    phones.append(ph)
+                    tones.append(tn)
+                else:
+                    phones.append(ph)
+                    tones.append(0)
+    # todo: implement word2ph
+    word2ph = [1 for i in phones]
+
+    phones = [post_replace_ph(i) for i in phones]
+    return phones, tones, word2ph
+
+def g2p(text, pad_start_end=True, tokenized=None):
+    if tokenized is None:
+        tokenized = tokenizer.tokenize(text)
+    # import pdb; pdb.set_trace()
+    phs = []
+    ph_groups = []
+    for t in tokenized:
+        if not t.startswith("#"):
+            ph_groups.append([t])
+        else:
+            ph_groups[-1].append(t.replace("#", ""))
+    
+    phones = []
+    tones = []
+    word2ph = []
+    for group in ph_groups:
+        w = "".join(group)
+        phone_len = 0
+        word_len = len(group)
+        if w.upper() in eng_dict:
+            phns, tns = refine_syllables(eng_dict[w.upper()])
+            phones += phns
+            tones += tns
+            phone_len += len(phns)
+        else:
+            phone_list = list(filter(lambda p: p != " ", _g2p(w)))
+            for ph in phone_list:
+                if ph in arpa:
+                    ph, tn = refine_ph(ph)
+                    phones.append(ph)
+                    tones.append(tn)
+                else:
+                    phones.append(ph)
+                    tones.append(0)
+                phone_len += 1
+        aaa = distribute_phone(phone_len, word_len)
+        word2ph += aaa
+    phones = [post_replace_ph(i) for i in phones]
+
+    if pad_start_end:
+        phones = ["_"] + phones + ["_"]
+        tones = [0] + tones + [0]
+        word2ph = [1] + word2ph + [1]
+    return phones, tones, word2ph
+
+def get_bert_feature(text, word2ph, device=None):
+    from text import english_bert
+
+    return english_bert.get_bert_feature(text, word2ph, device=device)
+
+if __name__ == "__main__":
+    # print(get_dict())
+    # print(eng_word_to_phoneme("hello"))
+    from text.english_bert import get_bert_feature
+    text = "In this paper, we propose 1 DSPGAN, a N-F-T GAN-based universal vocoder."
+    text = text_normalize(text)
+    phones, tones, word2ph = g2p(text)
+    import pdb; pdb.set_trace()
+    bert = get_bert_feature(text, word2ph)
+    
+    print(phones, tones, word2ph, bert.shape)
+
+    # all_phones = set()
+    # for k, syllables in eng_dict.items():
+    #     for group in syllables:
+    #         for ph in group:
+    #             all_phones.add(ph)
+    # print(all_phones)

xinference/thirdparty/melo/text/english_bert.py

@@ -0,0 +1,39 @@
+import torch
+from transformers import AutoTokenizer, AutoModelForMaskedLM
+import sys
+
+model_id = 'bert-base-uncased'
+tokenizer = AutoTokenizer.from_pretrained(model_id)
+model = None
+
+def get_bert_feature(text, word2ph, device=None):
+    global model
+    if (
+        sys.platform == "darwin"
+        and torch.backends.mps.is_available()
+        and device == "cpu"
+    ):
+        device = "mps"
+    if not device:
+        device = "cuda"
+    if model is None:
+        model = AutoModelForMaskedLM.from_pretrained(model_id).to(
+            device
+        )
+    with torch.no_grad():
+        inputs = tokenizer(text, return_tensors="pt")
+        for i in inputs:
+            inputs[i] = inputs[i].to(device)
+        res = model(**inputs, output_hidden_states=True)
+        res = torch.cat(res["hidden_states"][-3:-2], -1)[0].cpu()
+        
+    assert inputs["input_ids"].shape[-1] == len(word2ph)
+    word2phone = word2ph
+    phone_level_feature = []
+    for i in range(len(word2phone)):
+        repeat_feature = res[i].repeat(word2phone[i], 1)
+        phone_level_feature.append(repeat_feature)
+
+    phone_level_feature = torch.cat(phone_level_feature, dim=0)
+
+    return phone_level_feature.T

xinference/thirdparty/melo/text/english_utils/abbreviations.py

@@ -0,0 +1,35 @@
+import re
+
+# List of (regular expression, replacement) pairs for abbreviations in english:
+abbreviations_en = [
+    (re.compile("\\b%s\\." % x[0], re.IGNORECASE), x[1])
+    for x in [
+        ("mrs", "misess"),
+        ("mr", "mister"),
+        ("dr", "doctor"),
+        ("st", "saint"),
+        ("co", "company"),
+        ("jr", "junior"),
+        ("maj", "major"),
+        ("gen", "general"),
+        ("drs", "doctors"),
+        ("rev", "reverend"),
+        ("lt", "lieutenant"),
+        ("hon", "honorable"),
+        ("sgt", "sergeant"),
+        ("capt", "captain"),
+        ("esq", "esquire"),
+        ("ltd", "limited"),
+        ("col", "colonel"),
+        ("ft", "fort"),
+    ]
+]
+
+def expand_abbreviations(text, lang="en"):
+    if lang == "en":
+        _abbreviations = abbreviations_en
+    else:
+        raise NotImplementedError()
+    for regex, replacement in _abbreviations:
+        text = re.sub(regex, replacement, text)
+    return text

xinference/thirdparty/melo/text/english_utils/number_norm.py

@@ -0,0 +1,97 @@
+""" from https://github.com/keithito/tacotron """
+
+import re
+from typing import Dict
+
+import inflect
+
+_inflect = inflect.engine()
+_comma_number_re = re.compile(r"([0-9][0-9\,]+[0-9])")
+_decimal_number_re = re.compile(r"([0-9]+\.[0-9]+)")
+_currency_re = re.compile(r"(£|\$|¥)([0-9\,\.]*[0-9]+)")
+_ordinal_re = re.compile(r"[0-9]+(st|nd|rd|th)")
+_number_re = re.compile(r"-?[0-9]+")
+
+
+def _remove_commas(m):
+    return m.group(1).replace(",", "")
+
+
+def _expand_decimal_point(m):
+    return m.group(1).replace(".", " point ")
+
+
+def __expand_currency(value: str, inflection: Dict[float, str]) -> str:
+    parts = value.replace(",", "").split(".")
+    if len(parts) > 2:
+        return f"{value} {inflection[2]}"  # Unexpected format
+    text = []
+    integer = int(parts[0]) if parts[0] else 0
+    if integer > 0:
+        integer_unit = inflection.get(integer, inflection[2])
+        text.append(f"{integer} {integer_unit}")
+    fraction = int(parts[1]) if len(parts) > 1 and parts[1] else 0
+    if fraction > 0:
+        fraction_unit = inflection.get(fraction / 100, inflection[0.02])
+        text.append(f"{fraction} {fraction_unit}")
+    if len(text) == 0:
+        return f"zero {inflection[2]}"
+    return " ".join(text)
+
+
+def _expand_currency(m: "re.Match") -> str:
+    currencies = {
+        "$": {
+            0.01: "cent",
+            0.02: "cents",
+            1: "dollar",
+            2: "dollars",
+        },
+        "€": {
+            0.01: "cent",
+            0.02: "cents",
+            1: "euro",
+            2: "euros",
+        },
+        "£": {
+            0.01: "penny",
+            0.02: "pence",
+            1: "pound sterling",
+            2: "pounds sterling",
+        },
+        "¥": {
+            # TODO rin
+            0.02: "sen",
+            2: "yen",
+        },
+    }
+    unit = m.group(1)
+    currency = currencies[unit]
+    value = m.group(2)
+    return __expand_currency(value, currency)
+
+
+def _expand_ordinal(m):
+    return _inflect.number_to_words(m.group(0))
+
+
+def _expand_number(m):
+    num = int(m.group(0))
+    if 1000 < num < 3000:
+        if num == 2000:
+            return "two thousand"
+        if 2000 < num < 2010:
+            return "two thousand " + _inflect.number_to_words(num % 100)
+        if num % 100 == 0:
+            return _inflect.number_to_words(num // 100) + " hundred"
+        return _inflect.number_to_words(num, andword="", zero="oh", group=2).replace(", ", " ")
+    return _inflect.number_to_words(num, andword="")
+
+
+def normalize_numbers(text):
+    text = re.sub(_comma_number_re, _remove_commas, text)
+    text = re.sub(_currency_re, _expand_currency, text)
+    text = re.sub(_decimal_number_re, _expand_decimal_point, text)
+    text = re.sub(_ordinal_re, _expand_ordinal, text)
+    text = re.sub(_number_re, _expand_number, text)
+    return text

xinference/thirdparty/melo/text/english_utils/time_norm.py

@@ -0,0 +1,47 @@
+import re
+
+import inflect
+
+_inflect = inflect.engine()
+
+_time_re = re.compile(
+    r"""\b
+                          ((0?[0-9])|(1[0-1])|(1[2-9])|(2[0-3]))  # hours
+                          :
+                          ([0-5][0-9])                            # minutes
+                          \s*(a\\.m\\.|am|pm|p\\.m\\.|a\\.m|p\\.m)? # am/pm
+                          \b""",
+    re.IGNORECASE | re.X,
+)
+
+
+def _expand_num(n: int) -> str:
+    return _inflect.number_to_words(n)
+
+
+def _expand_time_english(match: "re.Match") -> str:
+    hour = int(match.group(1))
+    past_noon = hour >= 12
+    time = []
+    if hour > 12:
+        hour -= 12
+    elif hour == 0:
+        hour = 12
+        past_noon = True
+    time.append(_expand_num(hour))
+
+    minute = int(match.group(6))
+    if minute > 0:
+        if minute < 10:
+            time.append("oh")
+        time.append(_expand_num(minute))
+    am_pm = match.group(7)
+    if am_pm is None:
+        time.append("p m" if past_noon else "a m")
+    else:
+        time.extend(list(am_pm.replace(".", "")))
+    return " ".join(time)
+
+
+def expand_time_english(text: str) -> str:
+    return re.sub(_time_re, _expand_time_english, text)

xinference/thirdparty/melo/text/es_phonemizer/base.py

@@ -0,0 +1,140 @@
+import abc
+from typing import List, Tuple
+
+from .punctuation import Punctuation
+
+
+class BasePhonemizer(abc.ABC):
+    """Base phonemizer class
+
+    Phonemization follows the following steps:
+        1. Preprocessing:
+            - remove empty lines
+            - remove punctuation
+            - keep track of punctuation marks
+
+        2. Phonemization:
+            - convert text to phonemes
+
+        3. Postprocessing:
+            - join phonemes
+            - restore punctuation marks
+
+    Args:
+        language (str):
+            Language used by the phonemizer.
+
+        punctuations (List[str]):
+            List of punctuation marks to be preserved.
+
+        keep_puncs (bool):
+            Whether to preserve punctuation marks or not.
+    """
+
+    def __init__(self, language, punctuations=Punctuation.default_puncs(), keep_puncs=False):
+        # ensure the backend is installed on the system
+        if not self.is_available():
+            raise RuntimeError("{} not installed on your system".format(self.name()))  # pragma: nocover
+
+        # ensure the backend support the requested language
+        self._language = self._init_language(language)
+
+        # setup punctuation processing
+        self._keep_puncs = keep_puncs
+        self._punctuator = Punctuation(punctuations)
+
+    def _init_language(self, language):
+        """Language initialization
+
+        This method may be overloaded in child classes (see Segments backend)
+
+        """
+        if not self.is_supported_language(language):
+            raise RuntimeError(f'language "{language}" is not supported by the ' f"{self.name()} backend")
+        return language
+
+    @property
+    def language(self):
+        """The language code configured to be used for phonemization"""
+        return self._language
+
+    @staticmethod
+    @abc.abstractmethod
+    def name():
+        """The name of the backend"""
+        ...
+
+    @classmethod
+    @abc.abstractmethod
+    def is_available(cls):
+        """Returns True if the backend is installed, False otherwise"""
+        ...
+
+    @classmethod
+    @abc.abstractmethod
+    def version(cls):
+        """Return the backend version as a tuple (major, minor, patch)"""
+        ...
+
+    @staticmethod
+    @abc.abstractmethod
+    def supported_languages():
+        """Return a dict of language codes -> name supported by the backend"""
+        ...
+
+    def is_supported_language(self, language):
+        """Returns True if `language` is supported by the backend"""
+        return language in self.supported_languages()
+
+    @abc.abstractmethod
+    def _phonemize(self, text, separator):
+        """The main phonemization method"""
+
+    def _phonemize_preprocess(self, text) -> Tuple[List[str], List]:
+        """Preprocess the text before phonemization
+
+        1. remove spaces
+        2. remove punctuation
+
+        Override this if you need a different behaviour
+        """
+        text = text.strip()
+        if self._keep_puncs:
+            # a tuple (text, punctuation marks)
+            return self._punctuator.strip_to_restore(text)
+        return [self._punctuator.strip(text)], []
+
+    def _phonemize_postprocess(self, phonemized, punctuations) -> str:
+        """Postprocess the raw phonemized output
+
+        Override this if you need a different behaviour
+        """
+        if self._keep_puncs:
+            return self._punctuator.restore(phonemized, punctuations)[0]
+        return phonemized[0]
+
+    def phonemize(self, text: str, separator="|", language: str = None) -> str:  # pylint: disable=unused-argument
+        """Returns the `text` phonemized for the given language
+
+        Args:
+            text (str):
+                Text to be phonemized.
+
+            separator (str):
+                string separator used between phonemes. Default to '_'.
+
+        Returns:
+            (str): Phonemized text
+        """
+        text, punctuations = self._phonemize_preprocess(text)
+        phonemized = []
+        for t in text:
+            p = self._phonemize(t, separator)
+            phonemized.append(p)
+        phonemized = self._phonemize_postprocess(phonemized, punctuations)
+        return phonemized
+
+    def print_logs(self, level: int = 0):
+        indent = "\t" * level
+        print(f"{indent}| > phoneme language: {self.language}")
+        print(f"{indent}| > phoneme backend: {self.name()}")