Initial commit: FunASR Speech Recognition Toolkit

Add complete FunASR codebase including models, runtime, and documentation.

funasr/tokenizer/abs_tokenizer.py

@@ -0,0 +1,142 @@
+import json
+import numpy as np
+from abc import ABC
+from pathlib import Path
+from abc import abstractmethod
+from typing import Union, Iterable, List, Dict
+
+
+class AbsTokenizer(ABC):
+    @abstractmethod
+    def text2tokens(self, line: str) -> List[str]:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        raise NotImplementedError
+
+    @abstractmethod
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        raise NotImplementedError
+
+
+class BaseTokenizer(ABC):
+    def __init__(
+        self,
+        token_list: Union[Path, str, Iterable[str]] = None,
+        unk_symbol: str = "<unk>",
+        **kwargs,
+    ):
+
+        """Initialize BaseTokenizer.
+        
+            Args:
+                token_list: TODO.
+                unk_symbol: TODO.
+                **kwargs: Additional keyword arguments.
+            """
+        if token_list is not None:
+            if isinstance(token_list, (Path, str)) and token_list.endswith(".txt"):
+                token_list = Path(token_list)
+                self.token_list_repr = str(token_list)
+                self.token_list: List[str] = []
+
+                with token_list.open("r", encoding="utf-8") as f:
+                    for idx, line in enumerate(f):
+                        line = line.rstrip()
+                        self.token_list.append(line)
+            elif isinstance(token_list, (Path, str)) and token_list.endswith(".json"):
+                token_list = Path(token_list)
+                self.token_list_repr = str(token_list)
+                self.token_list: List[str] = []
+
+                with open(token_list, "r", encoding="utf-8") as f:
+                    self.token_list = json.load(f)
+
+            else:
+                self.token_list: List[str] = list(token_list)
+                self.token_list_repr = ""
+                for i, t in enumerate(self.token_list):
+                    if i == 3:
+                        break
+                    self.token_list_repr += f"{t}, "
+                self.token_list_repr += f"... (NVocab={(len(self.token_list))})"
+
+            self.token2id: Dict[str, int] = {}
+            for i, t in enumerate(self.token_list):
+                if t in self.token2id:
+                    raise RuntimeError(f'Symbol "{t}" is duplicated')
+                self.token2id[t] = i
+
+            self.unk_symbol = unk_symbol
+            if self.unk_symbol not in self.token2id:
+                raise RuntimeError(f"Unknown symbol '{unk_symbol}' doesn't exist in the token_list")
+            self.unk_id = self.token2id[self.unk_symbol]
+
+    def encode(self, text, **kwargs):
+        """Encode.
+        
+            Args:
+                text: Text tensor or string input.
+                **kwargs: Additional keyword arguments.
+            """
+        tokens = self.text2tokens(text)
+        text_ints = self.tokens2ids(tokens)
+
+        return text_ints
+
+    def decode(self, text_ints):
+        """Decode.
+        
+            Args:
+                text_ints: TODO.
+            """
+        token = self.ids2tokens(text_ints)
+        text = self.tokens2text(token)
+        return text
+
+    def get_num_vocabulary_size(self) -> int:
+        """Get num vocabulary size."""
+        return len(self.token_list)
+
+    def ids2tokens(self, integers: Union[np.ndarray, Iterable[int]]) -> List[str]:
+        """Ids2tokens.
+        
+            Args:
+                integers: TODO.
+            """
+        if isinstance(integers, np.ndarray) and integers.ndim != 1:
+            raise ValueError(f"Must be 1 dim ndarray, but got {integers.ndim}")
+        return [self.token_list[i] for i in integers]
+
+    def tokens2ids(self, tokens: Iterable[str]) -> List[int]:
+        """Tokens2ids.
+        
+            Args:
+                tokens: TODO.
+            """
+        return [self.token2id.get(i, self.unk_id) for i in tokens]
+
+    @abstractmethod
+    def text2tokens(self, line: str) -> List[str]:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        raise NotImplementedError
+
+    @abstractmethod
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        raise NotImplementedError

funasr/tokenizer/build_tokenizer.py

@@ -0,0 +1,59 @@
+from pathlib import Path
+from typing import Iterable
+from typing import Union
+
+
+from funasr.tokenizer.abs_tokenizer import AbsTokenizer
+from funasr.tokenizer.char_tokenizer import CharTokenizer
+from funasr.tokenizer.phoneme_tokenizer import PhonemeTokenizer
+from funasr.tokenizer.sentencepiece_tokenizer import SentencepiecesTokenizer
+from funasr.tokenizer.word_tokenizer import WordTokenizer
+
+
+def build_tokenizer(
+    token_type: str,
+    bpemodel: Union[Path, str, Iterable[str]] = None,
+    non_linguistic_symbols: Union[Path, str, Iterable[str]] = None,
+    remove_non_linguistic_symbols: bool = False,
+    space_symbol: str = "<space>",
+    delimiter: str = None,
+    g2p_type: str = None,
+) -> AbsTokenizer:
+    """A helper function to instantiate Tokenizer"""
+    if token_type == "bpe":
+        if bpemodel is None:
+            raise ValueError('bpemodel is required if token_type = "bpe"')
+
+        if remove_non_linguistic_symbols:
+            raise RuntimeError(
+                "remove_non_linguistic_symbols is not implemented for token_type=bpe"
+            )
+        return SentencepiecesTokenizer(bpemodel)
+
+    elif token_type == "word":
+        if remove_non_linguistic_symbols and non_linguistic_symbols is not None:
+            return WordTokenizer(
+                delimiter=delimiter,
+                non_linguistic_symbols=non_linguistic_symbols,
+                remove_non_linguistic_symbols=True,
+            )
+        else:
+            return WordTokenizer(delimiter=delimiter)
+
+    elif token_type == "char":
+        return CharTokenizer(
+            non_linguistic_symbols=non_linguistic_symbols,
+            space_symbol=space_symbol,
+            remove_non_linguistic_symbols=remove_non_linguistic_symbols,
+        )
+
+    elif token_type == "phn":
+        return PhonemeTokenizer(
+            g2p_type=g2p_type,
+            non_linguistic_symbols=non_linguistic_symbols,
+            space_symbol=space_symbol,
+            remove_non_linguistic_symbols=remove_non_linguistic_symbols,
+        )
+
+    else:
+        raise ValueError(f"token_mode must be one of bpe, word, char or phn: " f"{token_type}")

funasr/tokenizer/char_tokenizer.py

@@ -0,0 +1,143 @@
+from pathlib import Path
+from typing import Iterable
+from typing import List
+from typing import Union
+import warnings
+import re
+
+from funasr.tokenizer.abs_tokenizer import BaseTokenizer
+from funasr.register import tables
+
+
+@tables.register("tokenizer_classes", "CharTokenizer")
+class CharTokenizer(BaseTokenizer):
+    def __init__(
+        self,
+        non_linguistic_symbols: Union[Path, str, Iterable[str]] = None,
+        space_symbol: str = "<space>",
+        remove_non_linguistic_symbols: bool = False,
+        split_with_space: bool = False,
+        seg_dict: str = None,
+        **kwargs,
+    ):
+        """Initialize CharTokenizer.
+        
+            Args:
+                non_linguistic_symbols: TODO.
+                space_symbol: TODO.
+                remove_non_linguistic_symbols: TODO.
+                split_with_space: TODO.
+                seg_dict: TODO.
+                **kwargs: Additional keyword arguments.
+            """
+        super().__init__(**kwargs)
+        self.space_symbol = space_symbol
+        if non_linguistic_symbols is None:
+            self.non_linguistic_symbols = set()
+        elif isinstance(non_linguistic_symbols, (Path, str)):
+            non_linguistic_symbols = Path(non_linguistic_symbols)
+            try:
+                with non_linguistic_symbols.open("r", encoding="utf-8") as f:
+                    self.non_linguistic_symbols = set(line.rstrip() for line in f)
+            except FileNotFoundError:
+                warnings.warn(f"{non_linguistic_symbols} doesn't exist.")
+                self.non_linguistic_symbols = set()
+        else:
+            self.non_linguistic_symbols = set(non_linguistic_symbols)
+        self.remove_non_linguistic_symbols = remove_non_linguistic_symbols
+        self.split_with_space = split_with_space
+        self.seg_dict = None
+        seg_dict = seg_dict if seg_dict is not None else kwargs.get("seg_dict_file", None)
+        if seg_dict is not None:
+            self.seg_dict = load_seg_dict(seg_dict)
+
+    def __repr__(self):
+        """Internal: repr  ."""
+        return (
+            f"{self.__class__.__name__}("
+            f'space_symbol="{self.space_symbol}"'
+            f'non_linguistic_symbols="{self.non_linguistic_symbols}"'
+            f")"
+        )
+
+    def text2tokens(self, line: Union[str, list]) -> List[str]:
+        # if self.split_with_space:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        if self.seg_dict is not None:
+            tokens = line.strip().split(" ")
+            tokens = seg_tokenize(tokens, self.seg_dict)
+        else:
+            tokens = []
+            while len(line) != 0:
+                for w in self.non_linguistic_symbols:
+                    if line.startswith(w):
+                        if not self.remove_non_linguistic_symbols:
+                            tokens.append(line[: len(w)])
+                        line = line[len(w) :]
+                        break
+                else:
+                    t = line[0]
+                    if t == " ":
+                        # t = "<space>"
+                        line = line[1:]
+                        continue
+                    tokens.append(t)
+                    line = line[1:]
+        return tokens
+
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        tokens = [t if t != self.space_symbol else " " for t in tokens]
+        return "".join(tokens)
+
+
+def load_seg_dict(seg_dict_file):
+    """Load seg dict.
+    
+        Args:
+            seg_dict_file: TODO.
+        """
+    seg_dict = {}
+    assert isinstance(seg_dict_file, str)
+    with open(seg_dict_file, "r", encoding="utf8") as f:
+        lines = f.readlines()
+        for line in lines:
+            s = line.strip().split()
+            key = s[0]
+            value = s[1:]
+            seg_dict[key] = " ".join(value)
+    return seg_dict
+
+
+def seg_tokenize(txt, seg_dict):
+    # pattern = re.compile(r'^[\u4E00-\u9FA50-9]+$')
+    """Seg tokenize.
+    
+        Args:
+            txt: TODO.
+            seg_dict: TODO.
+        """
+    pattern = re.compile(r"([\u4E00-\u9FA5A-Za-z0-9])")
+    out_txt = ""
+    for word in txt:
+        word = word.lower()
+        if word in seg_dict:
+            out_txt += seg_dict[word] + " "
+        else:
+            if pattern.match(word):
+                for char in word:
+                    if char in seg_dict:
+                        out_txt += seg_dict[char] + " "
+                    else:
+                        out_txt += "<unk>" + " "
+            else:
+                out_txt += "<unk>" + " "
+    return out_txt.strip().split()

funasr/tokenizer/cleaner.py

@@ -0,0 +1,58 @@
+from typing import Collection
+
+from jaconv import jaconv
+
+# import tacotron_cleaner.cleaners
+
+try:
+    from vietnamese_cleaner import vietnamese_cleaners
+except ImportError:
+    vietnamese_cleaners = None
+
+
+class TextCleaner:
+    """Text cleaner.
+
+    Examples:
+        >>> cleaner = TextCleaner("tacotron")
+        >>> cleaner("(Hello-World);   &  jr. & dr.")
+        'HELLO WORLD, AND JUNIOR AND DOCTOR'
+
+    """
+
+    def __init__(self, cleaner_types: Collection[str] = None):
+
+        """Initialize TextCleaner.
+        
+            Args:
+                cleaner_types: TODO.
+            """
+        if cleaner_types is None:
+            self.cleaner_types = []
+        elif isinstance(cleaner_types, str):
+            self.cleaner_types = [cleaner_types]
+        else:
+            self.cleaner_types = list(cleaner_types)
+
+    def __call__(self, text: str) -> str:
+        """Internal: call  .
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        for t in self.cleaner_types:
+            if t == "tacotron":
+                # text = tacotron_cleaner.cleaners.custom_english_cleaners(text)
+                pass
+            elif t == "jaconv":
+                text = jaconv.normalize(text)
+            elif t == "vietnamese":
+                if vietnamese_cleaners is None:
+                    raise RuntimeError("Please install underthesea")
+                text = vietnamese_cleaners.vietnamese_cleaner(text)
+            elif t == "korean_cleaner":
+                text = KoreanCleaner.normalize_text(text)
+            else:
+                raise RuntimeError(f"Not supported: type={t}")
+
+        return text

funasr/tokenizer/hf_tokenizer.py

@@ -0,0 +1,22 @@
+from funasr.register import tables
+
+
+@tables.register("tokenizer_classes", "HuggingfaceTokenizer")
+def HuggingfaceTokenizer(init_param_path, **kwargs):
+    """Huggingfacetokenizer.
+    
+        Args:
+            init_param_path: TODO.
+            **kwargs: Additional keyword arguments.
+        """
+    try:
+        from transformers import AutoTokenizer
+    except Exception as e:
+        raise ImportError(
+            "HuggingfaceTokenizer requires 'transformers'. "
+            "Please install it with: pip install -U transformers"
+        ) from e
+
+    tokenizer = AutoTokenizer.from_pretrained(init_param_path)
+
+    return tokenizer

funasr/tokenizer/korean_cleaner.py

@@ -0,0 +1,89 @@
+# Referenced from https://github.com/hccho2/Tacotron-Wavenet-Vocoder-Korean
+
+import re
+
+
+class KoreanCleaner:
+    @classmethod
+    def _normalize_numbers(cls, text):
+        """Internal: normalize numbers.
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        number_to_kor = {
+            "0": "영",
+            "1": "일",
+            "2": "이",
+            "3": "삼",
+            "4": "사",
+            "5": "오",
+            "6": "육",
+            "7": "칠",
+            "8": "팔",
+            "9": "구",
+        }
+        new_text = "".join(
+            number_to_kor[char] if char in number_to_kor.keys() else char for char in text
+        )
+        return new_text
+
+    @classmethod
+    def _normalize_english_text(cls, text):
+        """Internal: normalize english text.
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        upper_alphabet_to_kor = {
+            "A": "에이",
+            "B": "비",
+            "C": "씨",
+            "D": "디",
+            "E": "이",
+            "F": "에프",
+            "G": "지",
+            "H": "에이치",
+            "I": "아이",
+            "J": "제이",
+            "K": "케이",
+            "L": "엘",
+            "M": "엠",
+            "N": "엔",
+            "O": "오",
+            "P": "피",
+            "Q": "큐",
+            "R": "알",
+            "S": "에스",
+            "T": "티",
+            "U": "유",
+            "V": "브이",
+            "W": "더블유",
+            "X": "엑스",
+            "Y": "와이",
+            "Z": "지",
+        }
+        new_text = re.sub("[a-z]+", lambda x: str.upper(x.group()), text)
+        new_text = "".join(
+            upper_alphabet_to_kor[char] if char in upper_alphabet_to_kor.keys() else char
+            for char in new_text
+        )
+
+        return new_text
+
+    @classmethod
+    def normalize_text(cls, text):
+        # stage 0 : text strip
+        """Normalize text.
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        text = text.strip()
+
+        # stage 1 : normalize numbers
+        text = cls._normalize_numbers(text)
+
+        # stage 2 : normalize english text
+        text = cls._normalize_english_text(text)
+        return text

funasr/tokenizer/phoneme_tokenizer.py

@@ -0,0 +1,644 @@
+import logging
+from pathlib import Path
+import re
+from typing import Iterable
+from typing import List
+from typing import Optional
+from typing import Union
+import warnings
+
+# import g2p_en
+import jamo
+
+from funasr.tokenizer.abs_tokenizer import AbsTokenizer
+
+
+g2p_classes = [
+    None,
+    "g2p_en",
+    "g2p_en_no_space",
+    "pyopenjtalk",
+    "pyopenjtalk_kana",
+    "pyopenjtalk_accent",
+    "pyopenjtalk_accent_with_pause",
+    "pyopenjtalk_prosody",
+    "pypinyin_g2p",
+    "pypinyin_g2p_phone",
+    "espeak_ng_arabic",
+    "espeak_ng_german",
+    "espeak_ng_french",
+    "espeak_ng_spanish",
+    "espeak_ng_russian",
+    "espeak_ng_greek",
+    "espeak_ng_finnish",
+    "espeak_ng_hungarian",
+    "espeak_ng_dutch",
+    "espeak_ng_english_us_vits",
+    "espeak_ng_hindi",
+    "g2pk",
+    "g2pk_no_space",
+    "korean_jaso",
+    "korean_jaso_no_space",
+]
+
+
+def split_by_space(text) -> List[str]:
+    """Split by space.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    if "   " in text:
+        text = text.replace("   ", " <space> ")
+        return [c.replace("<space>", " ") for c in text.split(" ")]
+    else:
+        return text.split(" ")
+
+
+def pyopenjtalk_g2p(text) -> List[str]:
+    """Pyopenjtalk g2p.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    import pyopenjtalk
+
+    # phones is a str object separated by space
+    phones = pyopenjtalk.g2p(text, kana=False)
+    phones = phones.split(" ")
+    return phones
+
+
+def pyopenjtalk_g2p_accent(text) -> List[str]:
+    """Pyopenjtalk g2p accent.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    import pyopenjtalk
+    import re
+
+    phones = []
+    for labels in pyopenjtalk.run_frontend(text)[1]:
+        p = re.findall(r"\-(.*?)\+.*?\/A:([0-9\-]+).*?\/F:.*?_([0-9]+)", labels)
+        if len(p) == 1:
+            phones += [p[0][0], p[0][2], p[0][1]]
+    return phones
+
+
+def pyopenjtalk_g2p_accent_with_pause(text) -> List[str]:
+    """Pyopenjtalk g2p accent with pause.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    import pyopenjtalk
+    import re
+
+    phones = []
+    for labels in pyopenjtalk.run_frontend(text)[1]:
+        if labels.split("-")[1].split("+")[0] == "pau":
+            phones += ["pau"]
+            continue
+        p = re.findall(r"\-(.*?)\+.*?\/A:([0-9\-]+).*?\/F:.*?_([0-9]+)", labels)
+        if len(p) == 1:
+            phones += [p[0][0], p[0][2], p[0][1]]
+    return phones
+
+
+def pyopenjtalk_g2p_kana(text) -> List[str]:
+    """Pyopenjtalk g2p kana.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    import pyopenjtalk
+
+    kanas = pyopenjtalk.g2p(text, kana=True)
+    return list(kanas)
+
+
+def pyopenjtalk_g2p_prosody(text: str, drop_unvoiced_vowels: bool = True) -> List[str]:
+    """Extract phoneme + prosoody symbol sequence from input full-context labels.
+
+    The algorithm is based on `Prosodic features control by symbols as input of
+    sequence-to-sequence acoustic modeling for neural TTS`_ with some r9y9's tweaks.
+
+    Args:
+        text (str): Input text.
+        drop_unvoiced_vowels (bool): whether to drop unvoiced vowels.
+
+    Returns:
+        List[str]: List of phoneme + prosody symbols.
+
+    Examples:
+        >>> from funasr.tokenizer.phoneme_tokenizer import pyopenjtalk_g2p_prosody
+        >>> pyopenjtalk_g2p_prosody("こんにちは。")
+        ['^', 'k', 'o', '[', 'N', 'n', 'i', 'ch', 'i', 'w', 'a', '$']
+
+    .. _`Prosodic features control by symbols as input of sequence-to-sequence acoustic
+        modeling for neural TTS`: https://doi.org/10.1587/transinf.2020EDP7104
+
+    """
+    import pyopenjtalk
+
+    labels = pyopenjtalk.run_frontend(text)[1]
+    N = len(labels)
+
+    phones = []
+    for n in range(N):
+        lab_curr = labels[n]
+
+        # current phoneme
+        p3 = re.search(r"\-(.*?)\+", lab_curr).group(1)
+
+        # deal unvoiced vowels as normal vowels
+        if drop_unvoiced_vowels and p3 in "AEIOU":
+            p3 = p3.lower()
+
+        # deal with sil at the beginning and the end of text
+        if p3 == "sil":
+            assert n == 0 or n == N - 1
+            if n == 0:
+                phones.append("^")
+            elif n == N - 1:
+                # check question form or not
+                e3 = _numeric_feature_by_regex(r"!(\d+)_", lab_curr)
+                if e3 == 0:
+                    phones.append("$")
+                elif e3 == 1:
+                    phones.append("?")
+            continue
+        elif p3 == "pau":
+            phones.append("_")
+            continue
+        else:
+            phones.append(p3)
+
+        # accent type and position info (forward or backward)
+        a1 = _numeric_feature_by_regex(r"/A:([0-9\-]+)\+", lab_curr)
+        a2 = _numeric_feature_by_regex(r"\+(\d+)\+", lab_curr)
+        a3 = _numeric_feature_by_regex(r"\+(\d+)/", lab_curr)
+
+        # number of mora in accent phrase
+        f1 = _numeric_feature_by_regex(r"/F:(\d+)_", lab_curr)
+
+        a2_next = _numeric_feature_by_regex(r"\+(\d+)\+", labels[n + 1])
+        # accent phrase border
+        if a3 == 1 and a2_next == 1 and p3 in "aeiouAEIOUNcl":
+            phones.append("#")
+        # pitch falling
+        elif a1 == 0 and a2_next == a2 + 1 and a2 != f1:
+            phones.append("]")
+        # pitch rising
+        elif a2 == 1 and a2_next == 2:
+            phones.append("[")
+
+    return phones
+
+
+def _numeric_feature_by_regex(regex, s):
+    """Internal: numeric feature by regex.
+    
+        Args:
+            regex: TODO.
+            s: TODO.
+        """
+    match = re.search(regex, s)
+    if match is None:
+        return -50
+    return int(match.group(1))
+
+
+def pypinyin_g2p(text) -> List[str]:
+    """Pypinyin g2p.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    from pypinyin import pinyin
+    from pypinyin import Style
+
+    phones = [phone[0] for phone in pinyin(text, style=Style.TONE3)]
+    return phones
+
+
+def pypinyin_g2p_phone(text) -> List[str]:
+    """Pypinyin g2p phone.
+    
+        Args:
+            text: Text tensor or string input.
+        """
+    from pypinyin import pinyin
+    from pypinyin import Style
+    from pypinyin.style._utils import get_finals
+    from pypinyin.style._utils import get_initials
+
+    phones = [
+        p
+        for phone in pinyin(text, style=Style.TONE3)
+        for p in [
+            get_initials(phone[0], strict=True),
+            get_finals(phone[0], strict=True),
+        ]
+        if len(p) != 0
+    ]
+    return phones
+
+
+class G2p_en:
+    """On behalf of g2p_en.G2p.
+
+    g2p_en.G2p isn't pickalable and it can't be copied to the other processes
+    via multiprocessing module.
+    As a workaround, g2p_en.G2p is instantiated upon calling this class.
+
+    """
+
+    def __init__(self, no_space: bool = False):
+        """Initialize G2p_en.
+        
+            Args:
+                no_space: TODO.
+            """
+        self.no_space = no_space
+        self.g2p = None
+
+    def __call__(self, text) -> List[str]:
+        """Internal: call  .
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        if self.g2p is None:
+            self.g2p = g2p_en.G2p()
+
+        phones = self.g2p(text)
+        if self.no_space:
+            # remove space which represents word serapater
+            phones = list(filter(lambda s: s != " ", phones))
+        return phones
+
+
+class G2pk:
+    """On behalf of g2pk.G2p.
+
+    g2pk.G2p isn't pickalable and it can't be copied to the other processes
+    via multiprocessing module.
+    As a workaround, g2pk.G2p is instantiated upon calling this class.
+
+    """
+
+    def __init__(self, descritive=False, group_vowels=False, to_syl=False, no_space=False):
+        """Initialize G2pk.
+        
+            Args:
+                descritive: TODO.
+                group_vowels: TODO.
+                to_syl: TODO.
+                no_space: TODO.
+            """
+        self.descritive = descritive
+        self.group_vowels = group_vowels
+        self.to_syl = to_syl
+        self.no_space = no_space
+        self.g2p = None
+
+    def __call__(self, text) -> List[str]:
+        """Internal: call  .
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        if self.g2p is None:
+            import g2pk
+
+            self.g2p = g2pk.G2p()
+
+        phones = list(
+            self.g2p(
+                text,
+                descriptive=self.descritive,
+                group_vowels=self.group_vowels,
+                to_syl=self.to_syl,
+            )
+        )
+        if self.no_space:
+            # remove space which represents word serapater
+            phones = list(filter(lambda s: s != " ", phones))
+        return phones
+
+
+class Jaso:
+    PUNC = "!'(),-.:;?"
+    SPACE = " "
+
+    JAMO_LEADS = "".join([chr(_) for _ in range(0x1100, 0x1113)])
+    JAMO_VOWELS = "".join([chr(_) for _ in range(0x1161, 0x1176)])
+    JAMO_TAILS = "".join([chr(_) for _ in range(0x11A8, 0x11C3)])
+
+    VALID_CHARS = JAMO_LEADS + JAMO_VOWELS + JAMO_TAILS + PUNC + SPACE
+
+    def __init__(self, space_symbol=" ", no_space=False):
+        """Initialize Jaso.
+        
+            Args:
+                space_symbol: TODO.
+                no_space: TODO.
+            """
+        self.space_symbol = space_symbol
+        self.no_space = no_space
+
+    def _text_to_jaso(self, line: str) -> List[str]:
+        """Internal: text to jaso.
+        
+            Args:
+                line: TODO.
+            """
+        jasos = list(jamo.hangul_to_jamo(line))
+        return jasos
+
+    def _remove_non_korean_characters(self, tokens):
+        """Internal: remove non korean characters.
+        
+            Args:
+                tokens: TODO.
+            """
+        new_tokens = [token for token in tokens if token in self.VALID_CHARS]
+        return new_tokens
+
+    def __call__(self, text) -> List[str]:
+        """Internal: call  .
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        graphemes = [x for x in self._text_to_jaso(text)]
+        graphemes = self._remove_non_korean_characters(graphemes)
+
+        if self.no_space:
+            graphemes = list(filter(lambda s: s != " ", graphemes))
+        else:
+            graphemes = [x if x != " " else self.space_symbol for x in graphemes]
+        return graphemes
+
+
+class Phonemizer:
+    """Phonemizer module for various languages.
+
+    This is wrapper module of https://github.com/bootphon/phonemizer.
+    You can define various g2p modules by specifying options for phonemizer.
+
+    See available options:
+        https://github.com/bootphon/phonemizer/blob/master/phonemizer/phonemize.py#L32
+
+    """
+
+    def __init__(
+        self,
+        backend,
+        word_separator: Optional[str] = None,
+        syllable_separator: Optional[str] = None,
+        phone_separator: Optional[str] = " ",
+        strip=False,
+        split_by_single_token: bool = False,
+        **phonemizer_kwargs,
+    ):
+        # delayed import
+        """Initialize Phonemizer.
+        
+            Args:
+                backend: TODO.
+                word_separator: TODO.
+                syllable_separator: TODO.
+                phone_separator: TODO.
+                strip: TODO.
+                split_by_single_token: TODO.
+                **phonemizer_kwargs: Additional keyword arguments.
+            """
+        from phonemizer.backend import BACKENDS
+        from phonemizer.separator import Separator
+
+        self.separator = Separator(
+            word=word_separator,
+            syllable=syllable_separator,
+            phone=phone_separator,
+        )
+
+        # define logger to suppress the warning in phonemizer
+        logger = logging.getLogger("phonemizer")
+        logger.setLevel(logging.ERROR)
+        self.phonemizer = BACKENDS[backend](
+            **phonemizer_kwargs,
+            logger=logger,
+        )
+        self.strip = strip
+        self.split_by_single_token = split_by_single_token
+
+    def __call__(self, text) -> List[str]:
+        """Internal: call  .
+        
+            Args:
+                text: Text tensor or string input.
+            """
+        tokens = self.phonemizer.phonemize(
+            [text],
+            separator=self.separator,
+            strip=self.strip,
+            njobs=1,
+        )[0]
+        if not self.split_by_single_token:
+            return tokens.split()
+        else:
+            # "a: ab" -> ["a", ":", "<space>",  "a", "b"]
+            # TODO(kan-bayashi): space replacement should be dealt in PhonemeTokenizer
+            return [c.replace(" ", "<space>") for c in tokens]
+
+
+class PhonemeTokenizer(AbsTokenizer):
+    def __init__(
+        self,
+        g2p_type: Union[None, str],
+        non_linguistic_symbols: Union[Path, str, Iterable[str]] = None,
+        space_symbol: str = "<space>",
+        remove_non_linguistic_symbols: bool = False,
+    ):
+        """Initialize PhonemeTokenizer.
+        
+            Args:
+                g2p_type: TODO.
+                non_linguistic_symbols: TODO.
+                space_symbol: TODO.
+                remove_non_linguistic_symbols: TODO.
+            """
+        if g2p_type is None:
+            self.g2p = split_by_space
+        elif g2p_type == "g2p_en":
+            self.g2p = G2p_en(no_space=False)
+        elif g2p_type == "g2p_en_no_space":
+            self.g2p = G2p_en(no_space=True)
+        elif g2p_type == "pyopenjtalk":
+            self.g2p = pyopenjtalk_g2p
+        elif g2p_type == "pyopenjtalk_kana":
+            self.g2p = pyopenjtalk_g2p_kana
+        elif g2p_type == "pyopenjtalk_accent":
+            self.g2p = pyopenjtalk_g2p_accent
+        elif g2p_type == "pyopenjtalk_accent_with_pause":
+            self.g2p = pyopenjtalk_g2p_accent_with_pause
+        elif g2p_type == "pyopenjtalk_prosody":
+            self.g2p = pyopenjtalk_g2p_prosody
+        elif g2p_type == "pypinyin_g2p":
+            self.g2p = pypinyin_g2p
+        elif g2p_type == "pypinyin_g2p_phone":
+            self.g2p = pypinyin_g2p_phone
+        elif g2p_type == "espeak_ng_arabic":
+            self.g2p = Phonemizer(
+                language="ar",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_german":
+            self.g2p = Phonemizer(
+                language="de",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_french":
+            self.g2p = Phonemizer(
+                language="fr-fr",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_spanish":
+            self.g2p = Phonemizer(
+                language="es",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_russian":
+            self.g2p = Phonemizer(
+                language="ru",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_greek":
+            self.g2p = Phonemizer(
+                language="el",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_finnish":
+            self.g2p = Phonemizer(
+                language="fi",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_hungarian":
+            self.g2p = Phonemizer(
+                language="hu",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_dutch":
+            self.g2p = Phonemizer(
+                language="nl",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "espeak_ng_hindi":
+            self.g2p = Phonemizer(
+                language="hi",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+            )
+        elif g2p_type == "g2pk":
+            self.g2p = G2pk(no_space=False)
+        elif g2p_type == "g2pk_no_space":
+            self.g2p = G2pk(no_space=True)
+        elif g2p_type == "espeak_ng_english_us_vits":
+            # VITS official implementation-like processing
+            # Reference: https://github.com/jaywalnut310/vits
+            self.g2p = Phonemizer(
+                language="en-us",
+                backend="espeak",
+                with_stress=True,
+                preserve_punctuation=True,
+                strip=True,
+                word_separator=" ",
+                phone_separator="",
+                split_by_single_token=True,
+            )
+        elif g2p_type == "korean_jaso":
+            self.g2p = Jaso(space_symbol=space_symbol, no_space=False)
+        elif g2p_type == "korean_jaso_no_space":
+            self.g2p = Jaso(no_space=True)
+        else:
+            raise NotImplementedError(f"Not supported: g2p_type={g2p_type}")
+
+        self.g2p_type = g2p_type
+        self.space_symbol = space_symbol
+        if non_linguistic_symbols is None:
+            self.non_linguistic_symbols = set()
+        elif isinstance(non_linguistic_symbols, (Path, str)):
+            non_linguistic_symbols = Path(non_linguistic_symbols)
+            try:
+                with non_linguistic_symbols.open("r", encoding="utf-8") as f:
+                    self.non_linguistic_symbols = set(line.rstrip() for line in f)
+            except FileNotFoundError:
+                warnings.warn(f"{non_linguistic_symbols} doesn't exist.")
+                self.non_linguistic_symbols = set()
+        else:
+            self.non_linguistic_symbols = set(non_linguistic_symbols)
+        self.remove_non_linguistic_symbols = remove_non_linguistic_symbols
+
+    def __repr__(self):
+        """Internal: repr  ."""
+        return (
+            f"{self.__class__.__name__}("
+            f'g2p_type="{self.g2p_type}", '
+            f'space_symbol="{self.space_symbol}", '
+            f'non_linguistic_symbols="{self.non_linguistic_symbols}"'
+            ")"
+        )
+
+    def text2tokens(self, line: str) -> List[str]:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        tokens = []
+        while len(line) != 0:
+            for w in self.non_linguistic_symbols:
+                if line.startswith(w):
+                    if not self.remove_non_linguistic_symbols:
+                        tokens.append(line[: len(w)])
+                    line = line[len(w) :]
+                    break
+            else:
+                t = line[0]
+                tokens.append(t)
+                line = line[1:]
+
+        line = "".join(tokens)
+        tokens = self.g2p(line)
+        return tokens
+
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        # phoneme type is not invertible
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        return "".join(tokens)

funasr/tokenizer/sentencepiece_tokenizer.py

@@ -0,0 +1,100 @@
+from pathlib import Path
+from typing import Iterable
+from typing import List
+from typing import Union
+
+import sentencepiece as spm
+
+from funasr.tokenizer.abs_tokenizer import BaseTokenizer
+from funasr.register import tables
+
+
+@tables.register("tokenizer_classes", "SentencepiecesTokenizer")
+class SentencepiecesTokenizer(BaseTokenizer):
+    def __init__(self, bpemodel: Union[Path, str], **kwargs):
+        """Initialize SentencepiecesTokenizer.
+        
+            Args:
+                bpemodel: TODO.
+                **kwargs: Additional keyword arguments.
+            """
+        super().__init__(**kwargs)
+        self.bpemodel = str(bpemodel)
+        # NOTE(kamo):
+        # Don't build SentencePieceProcessor in __init__()
+        # because it's not picklable and it may cause following error,
+        # "TypeError: can't pickle SwigPyObject objects",
+        # when giving it as argument of "multiprocessing.Process()".
+        self.sp = None
+        self._build_sentence_piece_processor()
+
+    def __repr__(self):
+        """Internal: repr  ."""
+        return f'{self.__class__.__name__}(model="{self.bpemodel}")'
+
+    def _build_sentence_piece_processor(self):
+        # Build SentencePieceProcessor lazily.
+        """Internal: build sentence piece processor."""
+        if self.sp is None:
+            self.sp = spm.SentencePieceProcessor()
+            self.sp.load(self.bpemodel)
+
+    def text2tokens(self, line: str) -> List[str]:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        self._build_sentence_piece_processor()
+        return self.sp.EncodeAsPieces(line)
+
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        self._build_sentence_piece_processor()
+        return self.sp.DecodePieces(list(tokens))
+
+    def encode(self, line: str, **kwargs) -> List[int]:
+        """Encode.
+        
+            Args:
+                line: TODO.
+                **kwargs: Additional keyword arguments.
+            """
+        self._build_sentence_piece_processor()
+        return self.sp.EncodeAsIds(line)
+
+    def decode(self, line: List[int], **kwargs):
+        """Decode.
+        
+            Args:
+                line: TODO.
+                **kwargs: Additional keyword arguments.
+            """
+        self._build_sentence_piece_processor()
+        return self.sp.DecodeIds(line)
+
+    def get_vocab_size(self):
+        """Get vocab size."""
+        return self.sp.GetPieceSize()
+
+    def ids2tokens(self, *args, **kwargs):
+        """Ids2tokens.
+        
+            Args:
+                *args: Variable positional arguments.
+                **kwargs: Additional keyword arguments.
+            """
+        return self.decode(*args, **kwargs)
+
+    def tokens2ids(self, *args, **kwargs):
+        """Tokens2ids.
+        
+            Args:
+                *args: Variable positional arguments.
+                **kwargs: Additional keyword arguments.
+            """
+        return self.encode(*args, **kwargs)

funasr/tokenizer/token_id_converter.py

@@ -0,0 +1,73 @@
+from pathlib import Path
+from typing import Dict
+from typing import Iterable
+from typing import List
+from typing import Union
+
+import numpy as np
+
+
+class TokenIDConverter:
+    def __init__(
+        self,
+        token_list: Union[Path, str, Iterable[str]],
+        unk_symbol: str = "<unk>",
+    ):
+
+        """Initialize TokenIDConverter.
+        
+            Args:
+                token_list: TODO.
+                unk_symbol: TODO.
+            """
+        if isinstance(token_list, (Path, str)):
+            token_list = Path(token_list)
+            self.token_list_repr = str(token_list)
+            self.token_list: List[str] = []
+
+            with token_list.open("r", encoding="utf-8") as f:
+                for idx, line in enumerate(f):
+                    line = line.rstrip()
+                    self.token_list.append(line)
+
+        else:
+            self.token_list: List[str] = list(token_list)
+            self.token_list_repr = ""
+            for i, t in enumerate(self.token_list):
+                if i == 3:
+                    break
+                self.token_list_repr += f"{t}, "
+            self.token_list_repr += f"... (NVocab={(len(self.token_list))})"
+
+        self.token2id: Dict[str, int] = {}
+        for i, t in enumerate(self.token_list):
+            if t in self.token2id:
+                raise RuntimeError(f'Symbol "{t}" is duplicated')
+            self.token2id[t] = i
+
+        self.unk_symbol = unk_symbol
+        if self.unk_symbol not in self.token2id:
+            raise RuntimeError(f"Unknown symbol '{unk_symbol}' doesn't exist in the token_list")
+        self.unk_id = self.token2id[self.unk_symbol]
+
+    def get_num_vocabulary_size(self) -> int:
+        """Get num vocabulary size."""
+        return len(self.token_list)
+
+    def ids2tokens(self, integers: Union[np.ndarray, Iterable[int]]) -> List[str]:
+        """Ids2tokens.
+        
+            Args:
+                integers: TODO.
+            """
+        if isinstance(integers, np.ndarray) and integers.ndim != 1:
+            raise ValueError(f"Must be 1 dim ndarray, but got {integers.ndim}")
+        return [self.token_list[i] for i in integers]
+
+    def tokens2ids(self, tokens: Iterable[str]) -> List[int]:
+        """Tokens2ids.
+        
+            Args:
+                tokens: TODO.
+            """
+        return [self.token2id.get(i, self.unk_id) for i in tokens]

funasr/tokenizer/whisper_tokenizer.py

@@ -0,0 +1,52 @@
+from funasr.register import tables
+
+
+@tables.register("tokenizer_classes", "WhisperTokenizer")
+def WhisperTokenizer(**kwargs):
+    """Whispertokenizer.
+    
+        Args:
+            **kwargs: Additional keyword arguments.
+        """
+    try:
+        from whisper.tokenizer import get_tokenizer
+    except:
+        print("Notice: If you want to use whisper, please `pip install -U openai-whisper`")
+
+    language = kwargs.get("language", None)
+    task = kwargs.get("task", "transcribe")
+    is_multilingual = kwargs.get("is_multilingual", True)
+    num_languages = kwargs.get("num_languages", 99)
+    tokenizer = get_tokenizer(
+        multilingual=is_multilingual,
+        num_languages=num_languages,
+        language=language,
+        task=task,
+    )
+
+    return tokenizer
+
+
+@tables.register("tokenizer_classes", "SenseVoiceTokenizer")
+def SenseVoiceTokenizer(**kwargs):
+    """Sensevoicetokenizer.
+    
+        Args:
+            **kwargs: Additional keyword arguments.
+        """
+    from funasr.models.sense_voice.whisper_lib.tokenizer import get_tokenizer
+
+    language = kwargs.get("language", None)
+    task = kwargs.get("task", None)
+    is_multilingual = kwargs.get("is_multilingual", True)
+    num_languages = kwargs.get("num_languages", 8749)
+    vocab_path = kwargs.get("vocab_path", None)
+    tokenizer = get_tokenizer(
+        multilingual=is_multilingual,
+        num_languages=num_languages,
+        language=language,
+        task=task,
+        vocab_path=vocab_path,
+    )
+
+    return tokenizer

funasr/tokenizer/word_tokenizer.py

@@ -0,0 +1,73 @@
+from pathlib import Path
+from typing import Iterable
+from typing import List
+from typing import Union
+import warnings
+
+
+from funasr.tokenizer.abs_tokenizer import AbsTokenizer
+
+
+class WordTokenizer(AbsTokenizer):
+    def __init__(
+        self,
+        delimiter: str = None,
+        non_linguistic_symbols: Union[Path, str, Iterable[str]] = None,
+        remove_non_linguistic_symbols: bool = False,
+    ):
+        """Initialize WordTokenizer.
+        
+            Args:
+                delimiter: TODO.
+                non_linguistic_symbols: TODO.
+                remove_non_linguistic_symbols: TODO.
+            """
+        self.delimiter = delimiter
+
+        if not remove_non_linguistic_symbols and non_linguistic_symbols is not None:
+            warnings.warn(
+                "non_linguistic_symbols is only used " "when remove_non_linguistic_symbols = True"
+            )
+
+        if non_linguistic_symbols is None:
+            self.non_linguistic_symbols = set()
+        elif isinstance(non_linguistic_symbols, (Path, str)):
+            non_linguistic_symbols = Path(non_linguistic_symbols)
+            try:
+                with non_linguistic_symbols.open("r", encoding="utf-8") as f:
+                    self.non_linguistic_symbols = set(line.rstrip() for line in f)
+            except FileNotFoundError:
+                warnings.warn(f"{non_linguistic_symbols} doesn't exist.")
+                self.non_linguistic_symbols = set()
+        else:
+            self.non_linguistic_symbols = set(non_linguistic_symbols)
+        self.remove_non_linguistic_symbols = remove_non_linguistic_symbols
+
+    def __repr__(self):
+        """Internal: repr  ."""
+        return f'{self.__class__.__name__}(delimiter="{self.delimiter}")'
+
+    def text2tokens(self, line: str) -> List[str]:
+        """Text2tokens.
+        
+            Args:
+                line: TODO.
+            """
+        tokens = []
+        for t in line.split(self.delimiter):
+            if self.remove_non_linguistic_symbols and t in self.non_linguistic_symbols:
+                continue
+            tokens.append(t)
+        return tokens
+
+    def tokens2text(self, tokens: Iterable[str]) -> str:
+        """Tokens2text.
+        
+            Args:
+                tokens: TODO.
+            """
+        if self.delimiter is None:
+            delimiter = " "
+        else:
+            delimiter = self.delimiter
+        return delimiter.join(tokens)