# Copyright (c) 2009 Carnegie Mellon University # # You may copy and modify this freely under the same terms as # Sphinx-III """Read/write Sphinx dictionary files This module reads and writes the text format dictionary files used by SphinxTrain, Sphinx-III, and PocketSphinx. """ __author__ = "David Huggins-Daines " __version__ = "$Revision $" from collections import defaultdict from itertools import chain import re import io def open(filename, *args, **kwargs): with io.open(filename) as fh: return S3Dict(fh, *args, **kwargs) altre = re.compile(r'(.*)\(([^\)]+)\)') tagre = re.compile(r':[^\(]+') def word_to_str(w): if w[1] == 1: return w[0] else: return "%s(%d)" % w def str_to_word(w): m = altre.match(w) if m: word, alt = m.groups() return word, int(alt) else: return w, 1 class S3Dict(dict): """ Class for reading / processing Sphinx format dictionary files. """ def __init__(self, infile=None, preserve_alts=False, strip_classtags=False): self.preserve_alts = preserve_alts self.strip_classtags = strip_classtags self.phoneset = defaultdict(int) self.maxalt = defaultdict(int) if infile is not None: self.read(infile) def __contains__(self, key): if not isinstance(key, tuple): word, alt = str_to_word(key) return dict.__contains__(self, (word, alt)) else: return dict.__contains__(self, key) def __getitem__(self, key): if not isinstance(key, tuple): word, alt = str_to_word(key) return self.get_alt_phones(word, alt) else: return self.get_alt_phones(*key) def __setitem__(self, key, val): if not isinstance(key, tuple): word, alt = str_to_word(key) return self.set_alt_phones(word, alt, val) else: w, p = key self.set_alt_phones(w, p, val) def read(self, fh): """ Read dictionary from a file. """ for line in fh: line = line.strip() # Skip any comment lines as in cmudict source if line.startswith('##'): continue if line.startswith(';;'): continue # And blank lines if line == "": continue spam = line.split() word = spam[0] if self.strip_classtags: word = self.tagre.sub("", word) phones = spam[1:] self[word] = phones def write(self, fh): """ Write dictionary to a file. """ wlist = list(self.keys()) wlist.sort() for k in wlist: word = word_to_str(k) fh.write("%s\t%s\n" % (word.encode('utf8'), " ".join(self[k]))) def get_phones(self, word): """ Get default pronunciation for word. If word is not present, KeyError will be raised. """ return dict.__getitem__(self, (word, 1)) def get_alt_phones(self, word, alt): """ Get alternate pronunciaition #alt for word. Alternate pronunciations are numbered from 1, where 1 is the default pronunciation. If word is not present, KeyError will be raised. If no alternate pronunciation alt exists, IndexError will be raised. """ if not dict.__contains__(self, (word, 1)): raise KeyError elif not dict.__contains__(self, (word, alt)): raise IndexError("Alternate pronunciation index %d does not exist" % alt) return dict.__getitem__(self, (word, alt)) def set_phones(self, word, phones): """ Set default pronunciation for word. """ dict.__setitem__(self, (word, 1), phones) self.maxalt[word] = 1 for ph in phones: self.phoneset[ph] += 1 # FIXME: should make a class for this def set_alt_phones(self, word, alt, phones): """ Set alternate pronunciaition #alt for word. If alt is greater than the maximum alternate pronunciation index plus one for this dictionary, IndexError will be raised. """ if alt > self.maxalt[word] + 1: raise IndexError("Alternate pronunciation index %d too high" % alt) dict.__setitem__(self, (word, alt), phones) self.maxalt[word] = max(alt, self.maxalt[word]) for ph in phones: self.phoneset[ph] += 1 def add_alt_phones(self, word, phones): """ Add a new alternate pronunciation for word. """ self.maxalt[word] += 1 dict.__setitem__(self, (word, self.maxalt[word]), phones) for ph in phones: self.phoneset[ph] += 1 def del_alt_phones(self, word, alt): """ Delete alternate pronunciation alt for word. If no such alternate pronunciation exists, IndexError will be raised. If this S3Dict was created with preserve_alts=True, the indices of remaining alternate pronunciations will be preserved (you can still this alternative index with set_alt_phones()). Otherwise, the remaining alternate pronunciations will be renumbered accordingly. """ if not dict.__contains__(self, (word, alt)): raise IndexError("Alternate pronunciation index %d does not exist" % alt) for ph in self[word, alt]: self.phoneset[ph] -= 1 if self.phoneset[ph] == 0: # FIXME: make a class del self.phoneset[ph] del self[word, alt] if alt == self.maxalt[word]: self.maxalt[word] -= 1 if not self.preserve_alts: alts = [] for i in range(1, self.maxalt[word] + 1): if (word, i) in self: alts.append(dict.__getitem__(self, (word, i))) del self[word, i] self.del_phones(word) for i, phones in enumerate(alts): dict.__setitem__(self, (word, i + 1), phones) self.maxalt[word] = len(alts) def del_phones(self, word): """ Delete all pronunciations for word. If you only wish to delete the default pronunciation (it is strongly suggested that you don't do this), use del_alt_phones(word, 1). """ for i in range(1, self.maxalt[word] + 1): if dict.__contains__(self, (word, i)): for ph in self[word, i]: self.phoneset[ph] -= 1 if self.phoneset[ph] == 0: # FIXME: make a class del self.phoneset[ph] dict.__delitem__(self, (word, i)) del self.maxalt[word] def words(self): """ Iterate over base words in this dictionary. """ for word, alt in self: if alt == 1: yield word def alts(self, word): """ Iterate over alternative pronunciations for a word. """ for i in range(1, self.maxalt[word] + 1): if (word, i) in self: yield self[word, i] def copy(self, other, w): """ Copy all pronunciations of w from other to self. """ for phones in other.alts(w): self.add_alt_phones(w, phones) def union(self, other): """ Compute the union of two dictionaries, returning the resulting dictionary. Lists of alternate pronunciations for words will be merged between the two dictionaries. The numeric identifiers of said alternates will not be preserved, however, the default pronunciation in the output is guaranteed to be the default pronunciation in self. """ newdict = self.__class__() sw = set(self.words()) ow = set(other.words()) # Simply copy words not shared between inputs for w in sw ^ ow: if w in self: copy(newdict, self, w) elif w in other: copy(newdict, other, w) # Merge pronunciations for all others for w in sw & ow: # Uniquify them prons = set() for phones in chain(self.alts(w), other.alts(w)): prons.add(tuple(phones)) # Set default pronunciation newdict[w] = self[w] prons.remove(tuple(self[w])) # Add all others in arbitrary order for phones in prons: newdict.add_alt_phones(w, list(phones)) return newdict def convert_to_40(d): badalts = [] for w in d.words(): baw = [] for i, p in enumerate(d.alts(w)): # Remove 'DX' outright because it is ambiguous if 'DX' in p: baw.append((w, i+1)) # Otherwise do AX -> AH, AXR -> ER else: j = 0 while j < len(p): if p[j] == 'AX': p[j] = 'AH' elif p[j] == 'IX': p[j] = 'IH' elif p[j] == 'AXR': p[j] = 'ER' elif p[j] == 'NX': p[j] = 'N' elif p[j] == 'TD': p[j] = 'T' elif p[j] == 'DD': p[j] = 'D' elif p[j] == 'TS': p[j:j+1] = ['T', 'S'] j += 1 elif p[j] == 'EN': p[j:j+1] = ['AH', 'N'] j += 1 elif p[j] == 'EM': p[j:j+1] = ['AH', 'M'] j += 1 j += 1 badalts.extend(baw[::-1]) for w, i in badalts: d.del_alt_phones(w, i)