# XPhoneBERT : A Pre-trained Multilingual Model for Phoneme Representations for Text-to-Speech XPhoneBERT is the first pre-trained multilingual model for phoneme representations for text-to-speech(TTS). XPhoneBERT has the same model architecture as BERT-base, trained using the RoBERTa pre-training approach on 330M phoneme-level sentences from nearly 100 languages and locales. Experimental results show that employing XPhoneBERT as an input phoneme encoder significantly boosts the performance of a strong neural TTS model in terms of naturalness and prosody and also helps produce fairly high-quality speech with limited training data. The general architecture and experimental results of XPhoneBERT can be found in [our INTERSPEECH 2023 paper](https://www.doi.org/10.21437/Interspeech.2023-444): @inproceedings{xphonebert, title = {{XPhoneBERT: A Pre-trained Multilingual Model for Phoneme Representations for Text-to-Speech}}, author = {Linh The Nguyen and Thinh Pham and Dat Quoc Nguyen}, booktitle = {Proceedings of the 24th Annual Conference of the International Speech Communication Association (INTERSPEECH)}, year = {2023}, pages = {5506--5510} } **Please CITE** our paper when XPhoneBERT is used to help produce published results or is incorporated into other software. For further information or requests, please go to [XPhoneBERT's homepage](https://github.com/VinAIResearch/XPhoneBERT)! ## Using XPhoneBERT with `transformers` ### Installation - Install `transformers` with pip: `pip install transformers`, or install `transformers` [from source](https://huggingface.co/docs/transformers/installation#installing-from-source). - Install `text2phonemesequence`: `pip install text2phonemesequence`
Our [`text2phonemesequence`](https://github.com/thelinhbkhn2014/Text2PhonemeSequence) package is to convert text sequences into phoneme-level sequences, employed to construct our multilingual phoneme-level pre-training data. We build `text2phonemesequence` by incorporating the [CharsiuG2P](https://github.com/lingjzhu/CharsiuG2P/tree/main) and the [segments](https://pypi.org/project/segments/) toolkits that perform text-to-phoneme conversion and phoneme segmentation, respectively. - **Notes** - Initializing `text2phonemesequence` for each language requires its corresponding ISO 639-3 code. The ISO 639-3 codes of supported languages are available at [HERE](https://github.com/VinAIResearch/XPhoneBERT/blob/main/LanguageISO639-3Codes.md). - `text2phonemesequence` takes a word-segmented sequence as input. And users might also perform text normalization on the word-segmented sequence before feeding into `text2phonemesequence`. When creating our pre-training data, we perform word and sentence segmentation on all text documents in each language by using the [spaCy](https://spacy.io) toolkit, except for Vietnamese where we employ the [VnCoreNLP](https://github.com/vncorenlp/VnCoreNLP) toolkit. We also use the text normalization component from the [NVIDIA NeMo toolkit](https://github.com/NVIDIA/NeMo) for English, German, Spanish and Chinese, and the [Vinorm](https://github.com/v-nhandt21/Vinorm) text normalization package for Vietnamese. ### Pre-trained model Model | #params | Arch. | Max length | Pre-training data ---|---|---|---|--- `vinai/xphonebert-base` | 88M | base | 512 | 330M phoneme-level sentences from nearly 100 languages and locales ### Example usage ```python from transformers import AutoModel, AutoTokenizer from text2phonemesequence import Text2PhonemeSequence # Load XPhoneBERT model and its tokenizer xphonebert = AutoModel.from_pretrained("vinai/xphonebert-base") tokenizer = AutoTokenizer.from_pretrained("vinai/xphonebert-base") # Load Text2PhonemeSequence # text2phone_model = Text2PhonemeSequence(language='eng-us', is_cuda=True) text2phone_model = Text2PhonemeSequence(language='jpn', is_cuda=True) # Input sequence that is already WORD-SEGMENTED (and text-normalized if applicable) # sentence = "That is , it is a testing text ." sentence = "これ は 、 テスト テキスト です ." input_phonemes = text2phone_model.infer_sentence(sentence) input_ids = tokenizer(input_phonemes, return_tensors="pt") with torch.no_grad(): features = xphonebert(**input_ids) ```