PyThaiTTS

Open Source Thai Text-to-speech library in Python

class pythaitts.TTS(pretrained='lunarlist_onnx', mode='last_checkpoint', version='1.0', device: str = 'cpu')[source]

__init__(pretrained='lunarlist_onnx', mode='last_checkpoint', version='1.0', device: str = 'cpu') → None[source]

Parameters:

pretrained (str) – TTS pretrained (lunarlist_onnx, khanomtan, lunarlist, vachana)
mode (str) – pretrained mode (lunarlist_onnx and vachana don’t support)
version (str) – model version (default is 1.0 or 1.1)
device (str) – device for running model. (lunarlist_onnx and vachana support CPU only.)

Options for mode

last_checkpoint (default) - last checkpoint of model
best_model - Best model (best loss)

You can see more about khanomtan tts at https://github.com/wannaphong/KhanomTan-TTS-v1.0 and https://github.com/wannaphong/KhanomTan-TTS-v1.1 If you want to use khanomtan tts, you must to install coqui-tts before use the model by pip install coqui-tts.

For lunarlist tts model, you must to install nemo before use the model by pip install nemo_toolkit[‘tts’]. You can see more about lunarlist tts at https://link.medium.com/OpPjQis6wBb

For lunarlist_onnx tts model, You can see more about lunarlist tts at https://github.com/PyThaiNLP/thaitts-onnx

For vachana tts model, You can see more about vachana tts at https://github.com/VYNCX/VachanaTTS2

load_pretrained(version)[source]: Load pretrained

tts(text: str, speaker_idx: str = 'Linda', language_idx: str = 'th-th', return_type: str = 'file', filename: str | None = None, preprocess: bool = True)[source]

speech synthesis

Parameters:

text (str) – text
speaker_idx (str) – speaker (default is Linda for khanomtan, th_f_1 for vachana)
language_idx (str) – language (default is th-th)
return_type (str) – return type (default is file)
filename (str) – path filename for save wav file if return_type is file.
preprocess (bool) – whether to preprocess text (convert numbers to Thai text and expand ๆ). Default is True.

Text Preprocessing

PyThaiTTS provides text preprocessing functions to improve TTS quality.

pythaitts.preprocess_text(text: str, expand_numbers: bool = True, expand_maiyamok_char: bool = True) → str[source]

Preprocess Thai text for TTS by converting numbers to text and expanding ๆ.

Parameters:

text (str) – Input text to preprocess
expand_numbers (bool) – Whether to convert numbers to Thai text (default: True)
expand_maiyamok_char (bool) – Whether to expand ๆ character (default: True)

Returns:

Preprocessed text

Return type:

str

Examples:

>>> preprocess_text("ฉันมี 123 บาท")
'ฉันมี หนึ่งร้อยยี่สิบสาม บาท'
>>> preprocess_text("ดีๆ")
'ดีดี'
>>> preprocess_text("มี 5 คนๆ")
'มี ห้า คนคน'

pythaitts.num_to_thai(num_str: str) → str[source]

Convert number string to Thai text. Supports integers and decimals.

Parameters:: num_str (str) – Number string to convert (e.g., “123”, “1234”, “12.5”)
Returns:: Thai text representation
Return type:: str

Examples:

>>> num_to_thai("0")
'ศูนย์'
>>> num_to_thai("123")
'หนึ่งร้อยยี่สิบสาม'
>>> num_to_thai("1000")
'หนึ่งพัน'

pythaitts.expand_maiyamok(text: str) → str[source]

Expand Thai repetition character (ๆ) by repeating the previous word or syllable.

The mai yamok (ๆ) is a Thai repetition mark that indicates the previous word or syllable should be repeated.

Parameters:: text (str) – Text containing ๆ character
Returns:: Text with ๆ expanded
Return type:: str

Examples:

>>> expand_maiyamok("ช้าๆ")
'ช้าช้า'
>>> expand_maiyamok("ดีๆ")
'ดีดี'