Stage 1: rule-based split (separate character, punctuation and digit)

• hello, how are you -> hello | , | how are you

Stage 2: over-split text to substrings by budoux for Chinese mix with Japanese, (space) for not scripta continua

• 你喜欢看アニメ吗 -> 你 | 喜欢 | 看 | アニメ | 吗
• 昨天見た映画はとても感動的でした -> 昨天 | 見た | 映画 | は | とても | 感動 | 的 | で | した
• how are you -> how | are | you

Stage 3: concatenate substrings based on their languages using fast-langdetect, wordfreq and regex (rule-based)

• 你 | 喜欢 | 看 | アニメ | 吗 -> 你喜欢看 | アニメ | 吗
• 昨天 | 見た | 映画 | は | とても | 感動 | 的 | で | した -> 昨天 | 見た映画はとても感動的でした
• how | are | you -> how are you

• TTS (Text-To-Speech) model often fails on multi-language speech generation, there are two ways to do:
  • Train a model can pronounce multiple languages
  • (This Package) Separate sentence based on language first, then use different language models
• Existed models in NLP toolkit (e.g. SpaCy, jieba) is usually helpful for dealing with text in ONE language for each model. Which means multi-language texts need pre-process, like texts below:

你喜欢看アニメ吗？
Vielen Dank merci beaucoup for your help.
你最近好吗、最近どうですか？요즘 어떻게 지내요？sky is clear and sunny。

• 1. 💡How it works
• 2. 🪨Motivation
• 3. 📕Usage
  • 3.1. 🚀Installation
  • 3.2. Basic
    • 3.2.1. split_by_lang
    • 3.2.2. merge_across_digit
  • 3.3. Advanced
    • 3.3.1. usage of lang_map and default_lang (for your languages)
• 4. Acknowledgement
• 5. ✨Star History

You can install the package using pip:

pip install split-lang

from split_lang import LangSplitter
lang_splitter = LangSplitter()
text = "你喜欢看アニメ吗"

substr = lang_splitter.split_by_lang(
    text=text,
)
for index, item in enumerate(substr):
    print(f"{index}|{item.lang}:{item.text}")

0|zh:你喜欢看
1|ja:アニメ
2|zh:吗

from split_lang import LangSplitter
lang_splitter = LangSplitter(merge_across_punctuation=True)
import time
texts = [
    "你喜欢看アニメ吗？我也喜欢看",
    "Please star this project on GitHub, Thanks you. I love you请加星这个项目，谢谢你。我爱你この項目をスターしてください、ありがとうございます！愛してる",
]
time1 = time.time()
for text in texts:
    substr = lang_splitter.split_by_lang(
        text=text,
    )
    for index, item in enumerate(substr):
        print(f"{index}|{item.lang}:{item.text}")
    print("----------------------")
time2 = time.time()
print(time2 - time1)

0|zh:你喜欢看
1|ja:アニメ
2|zh:吗？我也喜欢看
----------------------
0|en:Please star this project on GitHub, Thanks you. I love you
1|zh:请加星这个项目，谢谢你。我爱你
2|ja:この項目をスターしてください、ありがとうございます！愛してる
----------------------
0.007998466491699219

lang_splitter.merge_across_digit = False
texts = [
    "衬衫的价格是9.15便士",
]
for text in texts:
    substr = lang_splitter.split_by_lang(
        text=text,
    )
    for index, item in enumerate(substr):
        print(f"{index}|{item.lang}:{item.text}")

0|zh:衬衫的价格是
1|digit:9.15
2|zh:便士

• default lang_map looks like below
  • if langua-py or fasttext or any other language detector detect the language that is NOT included in lang_map will be set to default_lang
  • if you set default_lang or value of key:value in lang_map to x, this substring will be merged to the near substring
    • zh | x | jp -> zh | jp (x been merged to one side)
    • In example below, zh-tw is set to x because character in zh and jp sometimes been detected as Traditional Chinese
• default default_lang is x

DEFAULT_LANG_MAP = {
    "zh": "zh",
    "yue": "zh",  # 粤语
    "wuu": "zh",  # 吴语
    "zh-cn": "zh",
    "zh-tw": "x",
    "ko": "ko",
    "ja": "ja",
    "de": "de",
    "fr": "fr",
    "en": "en",
    "hr": "en",
}
DEFAULT_LANG = "x"

• Inspired by LlmKira/fast-langdetect
• Text segmentation depends on google/budoux
• Language detection depends on zafercavdar/fasttext-langdetect and rspeer/wordfreq