Spaces:
Running on Zero
Running on Zero
| """Forced-alignment DP and dictionary helpers (TensorFlow-free).""" | |
| from __future__ import annotations | |
| import re | |
| import sys | |
| from pathlib import Path | |
| import numpy as np | |
| from maps_torch.phones import num2phn, phn2num | |
| from utils import align, collapse, fold_phone | |
| __all__ = [ | |
| "align", | |
| "collapse", | |
| "fold_phone", | |
| "load_dictionary", | |
| "force_align", | |
| "PhoneLabel", | |
| ] | |
| class PhoneLabel: | |
| def __init__(self, phone: str, duration: int): | |
| self.phone = phone | |
| self.duration = duration | |
| def __str__(self) -> str: | |
| return str([self.phone, self.duration]) | |
| def load_dictionary(dname: Path | str, cmuformat: bool = True) -> dict: | |
| mapping: dict = {} | |
| if not cmuformat: | |
| print("Custom dictionary formats not supported yet. Please format use CMU dict formatting and run again.") | |
| sys.exit(1) | |
| with open(dname, "r") as d: | |
| for line in d: | |
| line = line.strip() | |
| if not line or line.startswith(";;;"): | |
| continue | |
| all_items = line.split() | |
| if len(all_items) < 2: | |
| continue | |
| word = re.sub(r"\(\d*\)", "", all_items[0]) | |
| # NLTK cmudict uses ``WORD <variant_index> PHONE ...``; skip the index. | |
| pron_start = 2 if len(all_items) >= 3 and all_items[1].isdigit() else 1 | |
| pronunciation = [fold_phone(p) for p in all_items[pron_start:]] | |
| if not pronunciation: | |
| continue | |
| if word not in mapping: | |
| mapping[word] = [pronunciation] | |
| else: | |
| mapping[word].append(pronunciation) | |
| mapping["sil"] = [["H#"], []] | |
| return mapping | |
| def force_align(collapsed: list[str], yhat: np.ndarray) -> tuple[list[PhoneLabel], np.ndarray]: | |
| yhat = np.squeeze(yhat, 0) | |
| # Clamp only to keep log defined; MAPS uses abs(log(p)). | |
| yhat = np.clip(yhat, 1e-12, 1.0) | |
| predictions = np.abs(np.log(yhat)) | |
| collapsed_ids = [phn2num[x.lower()] for x in collapsed] | |
| a, m = align(collapsed_ids, predictions) | |
| a = [num2phn[p] for p in a] | |
| seq = [PhoneLabel(phone=a[0], duration=1)] | |
| for elem in a[1:]: | |
| if seq[-1].phone != elem: | |
| seq.append(PhoneLabel(phone=elem, duration=1)) | |
| else: | |
| seq[-1].duration += 1 | |
| return seq, m | |