"""Class definitions for Speakers, Files, Utterances and Jobs"""
from __future__ import annotations
import os
import typing
import unicodedata
from typing import TYPE_CHECKING, Optional, Union
from kalpy.gmm.data import CtmInterval
from praatio import textgrid
from montreal_forced_aligner.corpus.helper import get_wav_info, load_text
from montreal_forced_aligner.data import SoundFileInformation, TextFileType
from montreal_forced_aligner.exceptions import TextParseError
from montreal_forced_aligner.textgrid import load_textgrid
if TYPE_CHECKING:
from dataclasses import dataclass
else:
from dataclassy import dataclass
__all__ = ["FileData", "UtteranceData"]
[docs]
@dataclass(slots=True)
class FileData:
"""
Data class for file information
Parameters
----------
name: str
File name
wav_path: str, optional
Path to sound file
text_path: str, optional
Path to sound file
relative_path: str
Path relative to corpus root directory
wav_info: dict[str, Any]
Information dictionary about the sound file
speaker_ordering: list[str]
List of speakers in the file
utterances: list[:class:`~montreal_forced_aligner.corpus.classes.UtteranceData`]
Utterance data for the file
"""
name: str
wav_path: typing.Optional[str]
text_path: typing.Optional[str]
text_type: TextFileType
relative_path: str
wav_info: SoundFileInformation = None
speaker_ordering: typing.List[str] = []
utterances: typing.List[UtteranceData] = []
[docs]
@classmethod
def parse_file(
cls,
file_name: str,
wav_path: Optional[str],
text_path: Optional[str],
relative_path: str,
speaker_characters: Union[int, str],
):
"""
Parse a collection of sound file and transcription file into a File
Parameters
----------
file_name: str
File identifier
wav_path: str
Full sound file path
text_path: str
Full transcription path
relative_path: str
Relative path from the corpus directory root
speaker_characters: int, optional
Number of characters in the file name to specify the speaker
Returns
-------
:class:`~montreal_forced_aligner.corpus.classes.FileData`
Parsed file
"""
text_type: TextFileType = TextFileType.NONE
if text_path is not None:
if str(text_path).lower().endswith(".textgrid"):
text_type = TextFileType.TEXTGRID
else:
text_type = TextFileType.LAB
file = FileData(
file_name, wav_path, text_path, relative_path=relative_path, text_type=text_type
)
if wav_path is not None:
root = os.path.dirname(wav_path)
file.wav_info = get_wav_info(wav_path)
else:
root = os.path.dirname(text_path)
if not speaker_characters:
speaker_name = os.path.basename(root)
elif isinstance(speaker_characters, int):
speaker_name = file_name[:speaker_characters]
elif speaker_characters == "prosodylab":
speaker_name = file_name.split("_")[1]
else:
speaker_name = file_name
file.load_text(speaker_name, ensure_root_speaker=bool(speaker_characters))
return file
[docs]
def load_text(self, root_speaker: str, ensure_root_speaker: bool = False) -> None:
"""
Load the transcription text from the text_file of the object
Parameters
----------
root_speaker: str
Speaker derived from the root directory, ignored for TextGrid tiers that have a speaker label
ensure_root_speaker: bool, defaults to False
Flag for ensuring that root_speaker is used as the speaker name
"""
if self.text_type == TextFileType.LAB:
try:
text = load_text(self.text_path)
text = unicodedata.normalize("NFKC", text)
except UnicodeDecodeError:
raise TextParseError(self.text_path)
if self.wav_info is None:
end = -1
else:
end = self.wav_info.duration
utterance = UtteranceData(
speaker_name=root_speaker,
file_name=self.name,
text=text,
begin=0,
channel=0,
end=end,
)
self.utterances.append(utterance)
self.speaker_ordering.append(root_speaker)
elif self.text_type == TextFileType.TEXTGRID:
tg = load_textgrid(self.text_path)
num_tiers = len(tg.tiers)
num_channels = 1
if self.wav_info is not None:
duration = self.wav_info.duration
num_channels = self.wav_info.num_channels
else:
duration = tg.maxTimestamp
if root_speaker:
self.speaker_ordering.append(root_speaker)
phone_data = {}
word_data = {}
for i, tier_name in enumerate(tg.tierNames):
if tier_name.endswith("phones"):
if " - " in tier_name:
speaker_name = tier_name.split(" - ")[0]
else:
speaker_name = root_speaker
if speaker_name not in phone_data:
phone_data[speaker_name] = []
ti = tg._tierDict[tier_name]
for begin, end, text in ti.entries:
text = text.strip()
if not text:
continue
begin, end = round(begin, 4), round(end, 4)
if begin >= duration:
continue
end = min(end, duration)
phone_data[speaker_name].append(CtmInterval(begin, end, text))
elif tier_name.endswith("words"):
if " - " in tier_name:
speaker_name = tier_name.split(" - ")[0]
else:
speaker_name = root_speaker
if speaker_name not in word_data:
word_data[speaker_name] = []
ti = tg._tierDict[tier_name]
for begin, end, text in ti.entries:
text = text.strip()
if not text:
continue
begin, end = round(begin, 4), round(end, 4)
if begin >= duration:
continue
end = min(end, duration)
word_data[speaker_name].append(CtmInterval(begin, end, text))
for i, tier_name in enumerate(tg.tierNames):
if tier_name.lower() == "notes":
continue
if tier_name.endswith("words") or tier_name.endswith("phones"):
if root_speaker not in self.speaker_ordering:
self.speaker_ordering.append(root_speaker)
continue
ti = tg._tierDict[tier_name]
if not isinstance(ti, textgrid.IntervalTier):
continue
if not ensure_root_speaker:
speaker_name = tier_name.strip()
else:
speaker_name = root_speaker
self.speaker_ordering.append(speaker_name)
channel = 0
if num_channels == 2 and i >= num_tiers / 2:
channel = 1
for begin, end, text in ti.entries:
text = text.strip()
if not text:
continue
text = unicodedata.normalize("NFKC", text)
begin, end = round(begin, 4), round(end, 4)
if begin >= duration:
continue
end = min(end, duration)
utt = UtteranceData(
speaker_name=speaker_name,
file_name=self.name,
begin=begin,
end=end,
text=text,
channel=channel,
)
if not utt.text:
continue
if speaker_name in phone_data:
for pi in phone_data[speaker_name]:
if pi.begin < utt.begin:
continue
if pi.end > utt.end:
break
if (
len(utt.phone_intervals) > 0
and pi.begin != utt.phone_intervals[-1].end
):
utt.phone_intervals.append(
CtmInterval(utt.phone_intervals[-1].end, pi.begin, "sil")
)
utt.phone_intervals.append(pi)
utt.manual_alignments = len(utt.phone_intervals) > 0
if utt.manual_alignments:
if utt.begin != utt.phone_intervals[0].begin:
if (
abs(utt.phone_intervals[0].begin - utt.begin) < 0.02
or utt.phone_intervals[0].label == "sil"
):
utt.phone_intervals[0].begin = utt.begin
else:
utt.phone_intervals.insert(
0,
CtmInterval(
utt.begin, utt.phone_intervals[0].begin, "sil"
),
)
if utt.phone_intervals[-1].end != utt.end:
if (
abs(utt.phone_intervals[-1].end - utt.end) < 0.02
or utt.phone_intervals[-1].label == "sil"
):
utt.phone_intervals[-1].end = utt.end
else:
utt.phone_intervals.append(
CtmInterval(utt.phone_intervals[-1].end, utt.end, "sil")
)
if speaker_name in word_data:
for wi in word_data[speaker_name]:
if wi.begin < utt.begin:
continue
if wi.end > utt.end:
break
if (
len(utt.word_intervals) > 0
and wi.begin != utt.word_intervals[-1].end
):
utt.word_intervals.append(
CtmInterval(utt.word_intervals[-1].end, wi.begin, "<eps>")
)
utt.word_intervals.append(wi)
if utt.manual_alignments and len(utt.word_intervals) > 0:
if utt.begin != utt.word_intervals[0].begin:
utt.word_intervals.insert(
0, CtmInterval(utt.begin, utt.word_intervals[0].begin, "<eps>")
)
if utt.word_intervals[-1].end != utt.end:
utt.word_intervals.append(
CtmInterval(utt.word_intervals[-1].end, utt.end, "<eps>")
)
self.utterances.append(utt)
else:
if self.wav_info is not None:
duration = self.wav_info.duration
else:
duration = 1
utt = UtteranceData(
speaker_name=root_speaker,
file_name=self.name,
begin=0,
channel=0,
end=duration,
)
self.utterances.append(utt)
self.speaker_ordering.append(root_speaker)
[docs]
@dataclass(slots=True)
class UtteranceData:
"""
Data class for utterance information
Parameters
----------
speaker_name: str
Speaker name
file_name: str
File name
begin: float, optional
Begin timestamp
end: float, optional
End timestamp
channel: int, optional
Sound file channel
text: str, optional
Utterance text
oovs: set[str]
Set of words not found in a look up
"""
speaker_name: str
file_name: str
begin: float
end: float
channel: int = 0
manual_alignments: bool = False
text: str = ""
normalized_text: str = ""
normalized_character_text: str = ""
oovs: str = ""
phone_intervals: typing.List = []
word_intervals: typing.List = []
