Spaces:
Running
Running
| # ================================================================================================== | |
| # DEEPFAKE AUDIO - synthesizer/utils/text.py (Linguistic Tokenization Engine) | |
| # ================================================================================================== | |
| # | |
| # 📝 DESCRIPTION | |
| # This module implements the text-to-sequence transformation logic. It handles | |
| # cleaning, ARPAbet embedding (via curly brace detection), and numeric ID | |
| # mapping, converting human-readable text into neural embeddings for the | |
| # synthesizer. | |
| # | |
| # 👤 AUTHORS | |
| # - Amey Thakur (https://github.com/Amey-Thakur) | |
| # - Mega Satish (https://github.com/msatmod) | |
| # | |
| # 🤝🏻 CREDITS | |
| # Original Real-Time Voice Cloning methodology by CorentinJ | |
| # Repository: https://github.com/CorentinJ/Real-Time-Voice-Cloning | |
| # | |
| # 🔗 PROJECT LINKS | |
| # Repository: https://github.com/Amey-Thakur/DEEPFAKE-AUDIO | |
| # Video Demo: https://youtu.be/i3wnBcbHDbs | |
| # Research: https://github.com/Amey-Thakur/DEEPFAKE-AUDIO/blob/main/DEEPFAKE-AUDIO.ipynb | |
| # | |
| # 📜 LICENSE | |
| # Released under the MIT License | |
| # Release Date: 2021-02-06 | |
| # ================================================================================================== | |
| from synthesizer.utils.symbols import symbols | |
| from synthesizer.utils import cleaners | |
| import re | |
| # Neural Mapping: Bidi dictionary for character/ID conversion | |
| _symbol_to_id = {s: i for i, s in enumerate(symbols)} | |
| _id_to_symbol = {i: s for i, s in enumerate(symbols)} | |
| # ARPAbet Detection: Identifies phonetic sequences in curly braces | |
| _curly_re = re.compile(r"(.*?)\{(.+?)\}(.*)") | |
| def text_to_sequence(text, cleaner_names): | |
| """ | |
| Symbolic Ingestion: | |
| Converts raw text into a categorical sequence of token IDs. | |
| Supports hybrid text/ARPAbet inputs via curly brace tagging. | |
| """ | |
| sequence = [] | |
| # Sequential Parsing Segment by Segment | |
| while len(text): | |
| m = _curly_re.match(text) | |
| if not m: | |
| sequence += _symbols_to_sequence(_clean_text(text, cleaner_names)) | |
| break | |
| sequence += _symbols_to_sequence(_clean_text(m.group(1), cleaner_names)) | |
| sequence += _arpabet_to_sequence(m.group(2)) | |
| text = m.group(3) | |
| # Termination: Append EOS token | |
| sequence.append(_symbol_to_id["~"]) | |
| return sequence | |
| def sequence_to_text(sequence): | |
| """Linguistic Restoration: Decodes token ID sequences back into human-readable text.""" | |
| result = "" | |
| for symbol_id in sequence: | |
| if symbol_id in _id_to_symbol: | |
| s = _id_to_symbol[symbol_id] | |
| # Handle ARPAbet re-bracketing | |
| if len(s) > 1 and s[0] == "@": | |
| s = "{%s}" % s[1:] | |
| result += s | |
| return result.replace("}{", " ") | |
| def _clean_text(text, cleaner_names): | |
| """Pipeline Orchestration: Runs text through specified normalization filters.""" | |
| for name in cleaner_names: | |
| cleaner = getattr(cleaners, name) | |
| if not cleaner: | |
| raise Exception("Unknown cleaner: %s" % name) | |
| text = cleaner(text) | |
| return text | |
| def _symbols_to_sequence(symbols): | |
| """ID Conversion: Maps a list of character tokens to their numeric analogues.""" | |
| return [_symbol_to_id[s] for s in symbols if _should_keep_symbol(s)] | |
| def _arpabet_to_sequence(text): | |
| """Phonetic Encoding: Converts space-delimited ARPAbet codes into ID sequences.""" | |
| return _symbols_to_sequence(["@" + s for s in text.split()]) | |
| def _should_keep_symbol(s): | |
| """Filter Logic: Ensures only valid tokens are included in the sequence.""" | |
| return s in _symbol_to_id and s not in ("_", "~") | |