Skip to content

Commit

Permalink
Use tokenizer for fast transcribe function.
Browse files Browse the repository at this point in the history
  • Loading branch information
@sultur
sultur committed Aug 28, 2023
1 parent 312a3ea commit a3a9055
Show file tree
Hide file tree
Showing 4 changed files with 188 additions and 49 deletions.
2 changes: 1 addition & 1 deletion README.md
View file
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
[![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://www.gnu.org/licenses/gpl-3.0)
[![Python 3.9](https://img.shields.io/badge/python-3.9-blue.svg)](https://www.python.org/downloads/release/python-390/)
[![tests](https://github.com/mideind/Icespeak/actions/workflows/main.yml/badge.svg)]()
[![License: GPL v3](https://img.shields.io/badge/License-GPLv3-blue.svg)](https://www.gnu.org/licenses/gpl-3.0)
[![pre-commit](https://img.shields.io/badge/pre--commit-enabled-brightgreen?logo=pre-commit)](https://github.com/pre-commit/pre-commit)
[![Ruff](https://img.shields.io/endpoint?url=https://raw.githubusercontent.com/astral-sh/ruff/main/assets/badge/v2.json)](https://github.com/astral-sh/ruff)

Expand Down
8 changes: 6 additions & 2 deletions src/icespeak/__init__.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,11 +20,15 @@
"""

from .parser import GreynirSSMLParser
from .transcribe import gssml
from .transcribe import DefaultTranscriber, gssml
from .tts import text_to_speech

fast_transcribe = DefaultTranscriber.token_transcribe

__all__ = (
"GreynirSSMLParser",
"text_to_speech",
"gssml"
"gssml",
"DefaultTranscriber",
"fast_transcribe",
)
2 changes: 1 addition & 1 deletion src/icespeak/settings.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -122,7 +122,7 @@ class Keys(BaseModel):

azure: Optional[AzureKey] = Field(default=None, description="Azure API key.")
aws: Optional[AWSPollyKey] = Field(default=None, description="AWS Polly API key.")
google: Optional[dict[Any,Any]] = Field(
google: Optional[dict[Any, Any]] = Field(
default=None, description="Path to Google API key file."
)

Expand Down
225 changes: 180 additions & 45 deletions src/icespeak/transcribe/__init__.py
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,11 +32,20 @@
from re import Match

from islenska.basics import ALL_CASES, ALL_GENDERS, ALL_NUMBERS
from reynir import TOK, Greynir, Tok
from reynir import Greynir
from reynir.bindb import GreynirBin
from reynir.simpletree import SimpleTree
from tokenizer import Abbreviations
from tokenizer.definitions import HYPHENS
from tokenizer import TOK, Abbreviations, Tok, detokenize, tokenize
from tokenizer.definitions import (
HYPHENS,
AmountTuple,
# CURRENCY_ABBREV,
# SI_UNITS,
# CURRENCY_SYMBOLS,
CurrencyTuple,
MeasurementTuple,
PunctuationTuple,
)

from .num import (
ROMAN_NUMERALS,
Expand Down Expand Up @@ -99,6 +108,23 @@ def gssml(data: Any = None, *, type: str, **kwargs: Union[str, float]) -> str:
)


def _currency_to_gender(code: str) -> GenderType:
non_kvk_currencies: Mapping[str, GenderType] = {
# KK
"USD": "kk",
"CHF": "kk",
"CAD": "kk",
# HK
"GBP": "hk",
"JPY": "hk",
"PLN": "hk",
"CNY": "hk",
"RMB": "hk",
"ZAR": "hk",
}
return non_kvk_currencies.get(code, "kvk")


# Spell out how character names are pronounced in Icelandic
_CHAR_PRONUNCIATION: Mapping[str, str] = {
"a": "a",
Expand Down Expand Up @@ -176,24 +202,50 @@ def gssml(data: Any = None, *, type: str, **kwargs: Union[str, float]) -> str:
"nóvember",
"desember",
)
_DATE_REGEXES = (
# Matches e.g. "1986-03-07"
re.compile(r"(?P<year>\d{1,4})-(?P<month>\d{1,2})-(?P<day>\d{1,2})"),
# Matches e.g. "1/4/2001"
re.compile(r"(?P<day>\d{1,2})/(?P<month>\d{1,2})/(?P<year>\d{1,4})"),
# Matches e.g. "25. janúar 1999" or "25 des."
re.compile(
r"(?P<day>\d{1,2})\.? ?"
+ r"(?P<month>(jan(úar|\.)?|feb(rúar|\.)?|mar(s|\.)?|"
+ r"apr(íl|\.)?|maí\.?|jún(í|\.)?|"
+ r"júl(í|\.)?|ágú(st|\.)?|sep(tember|\.)?|"
+ r"okt(óber|\.)?|nóv(ember|\.)?|des(ember|\.)?))" # 'month' capture group ends
+ r"( (?P<year>\d{1,4}))?", # Optional
flags=re.IGNORECASE,
_DATE_REGEX = re.compile(
"|".join(
( # TODO: This matches incorrect dates such as 1999-88-63 or 43/67/1999
# Matches e.g. "1986-03-07"
r"(?P<year1>\d{1,4})-(?P<month1>\d{1,2})-(?P<day1>\d{1,2})",
# Matches e.g. "1/4/2001"
r"(?P<day2>\d{1,2})/(?P<month2>\d{1,2})/(?P<year2>\d{1,4})",
# Matches e.g. "25. janúar 1999" or "25 des."
r"(?P<day3>\d{1,2})\.? ?"
+ r"(?P<month3>(jan(úar|\.)?|feb(rúar|\.)?|mar(s|\.)?|"
+ r"apr(íl|\.)?|maí\.?|jún(í|\.)?|"
+ r"júl(í|\.)?|ágú(st|\.)?|sept?(ember|\.)?|"
+ r"okt(óber|\.)?|nóv(ember|\.)?|des(ember|\.)?))" # 'month' capture group ends
+ r"( (?P<year3>\d{1,4}))?", # Optional
)
),
flags=re.IGNORECASE,
)


def _date_from_match(match: re.Match[str], case: CaseType = "nf") -> str:
gd = match.groupdict()

day_val: Optional[str] = next(
filter(bool, (gd[f"day{i}"] for i in range(1, 4))), None
)
month_val: Optional[str] = next(
filter(bool, (gd[f"month{i}"] for i in range(1, 4))), None
)
year_val: Optional[str] = next(
filter(bool, (gd[f"year{i}"] for i in range(1, 4))), None
)

day = number_to_ordinal(day_val, gender="kk", case=case, number="et")
mon: str = month_val
# Month names don't change in different declensions
month = (
_MONTH_NAMES[int(mon) - 1] # DD/MM/YYYY specification
if mon.isdecimal()
else _MONTH_NAMES[_MONTH_ABBREVS.index(mon[:3])] # Non-decimal
)
return f"{day} {month} {year_to_text(year_val)}" if year_val else f"{day} {month}"


def _split_substring_types(t: str) -> Iterable[str]:
"""
Split text into alphabetic, decimal or
Expand All @@ -203,11 +255,8 @@ def _split_substring_types(t: str) -> Iterable[str]:
list(_split_substring_types("hello world,123"))
-> ["hello", " ", "world", ",", "123"]
"""

def f(c: str) -> int:
return c.isalpha() + 2 * c.isdecimal()

return ("".join(g) for _, g in itertools.groupby(t, key=f))
chartype2val: Callable[[str], int] = lambda c: c.isalpha() + 2 * c.isdecimal()
return ("".join(g) for _, g in itertools.groupby(t, key=chartype2val))


# Matches letter followed by period or
Expand All @@ -221,7 +270,9 @@ def f(c: str) -> int:

# Terms common in sentences which refer to results from sports
_SPORTS_LEMMAS: frozenset[str] = frozenset(("leikur", "vinna", "tapa", "sigra"))

_IGNORED_TOKENS = frozenset(
(TOK.PUNCTUATION, TOK.WORD, TOK.PERSON, TOK.ENTITY, TOK.TIMESTAMP, TOK.UNKNOWN)
)
_HYPHEN_SYMBOLS = frozenset(HYPHENS)

_StrBool = Union[str, bool]
Expand Down Expand Up @@ -408,6 +459,7 @@ def digits(cls, txt: str) -> str:
@_empty_str
def phone(cls, txt: str) -> str:
"""Spell out a phone number."""
# TODO: "plús" for e.g. +354-5588-5522
return cls.digits(txt)

@classmethod
Expand Down Expand Up @@ -479,28 +531,13 @@ def _time_fmt(match: Match[str]) -> str:
@_empty_str
def date(cls, txt: str, case: CaseType = "nf") -> str:
"""Voicifies a date"""
for r in _DATE_REGEXES:
match = r.search(txt)
if match:
# Found match
start, end = match.span()
gd = match.groupdict()
day = number_to_ordinal(gd["day"], gender="kk", case=case, number="et")
mon: str = gd["month"]
# Month names don't change in different declensions
month = (
_MONTH_NAMES[int(mon) - 1] # DD/MM/YYYY specification
if mon.isdecimal()
else _MONTH_NAMES[_MONTH_ABBREVS.index(mon[:3])] # Non-decimal
)
fmt_date = (
f"{day} {month} {year_to_text(gd['year'])}"
if gd["year"]
else f"{day} {month}"
)
# Only replace date part, leave rest of string intact
txt = txt[:start] + fmt_date + txt[end:]
break
match = _DATE_REGEX.search(txt)
if match:
# Found match
start, end = match.span()
fmt_date = _date_from_match(match, case=case)
# Only replace date part, leave rest of string intact
txt = txt[:start] + fmt_date + txt[end:]
return txt

@classmethod
Expand Down Expand Up @@ -1087,3 +1124,101 @@ def paragraph(cls, txt: str) -> str:
def sentence(cls, txt: str) -> str:
"""Sentence delimiter for speech synthesis."""
return f"<s>{txt}</s>"

@classmethod
def token_transcribe(cls, text: str):
"""
Quick transcription of Icelandic text for TTS.
Utilizes the tokenizer library.
"""
tokens: list[Tok] = list(tokenize(text))
for token in tokens:
if token.kind in _IGNORED_TOKENS:
continue

elif token.kind == TOK.TIME:
token.txt = cls.time(token.txt)

elif token.kind == TOK.DATE:
token.txt = cls.date(token.txt, case="þf")

elif token.kind == TOK.YEAR:
token.txt = cls.year(token.integer)

elif token.kind == TOK.NUMBER:
token.txt = cls.float(token.number, case="nf", gender="hk")

elif token.kind == TOK.TELNO:
token.txt = cls.phone(token.txt)

elif token.kind == TOK.PERCENT:
pass # TODO

elif token.kind == TOK.URL:
protocol, _, domain = token.txt.partition("://")
if domain:
token.txt = cls.spell(protocol) + cls.domain(domain)

elif token.kind == TOK.ORDINAL:
token.txt = cls.ordinal(token.ordinal, case="nf", gender="kk")

elif token.kind == TOK.CURRENCY:
curr, _, _ = cast(CurrencyTuple, token.val)
token.txt = cls.currency(curr)

elif token.kind == TOK.AMOUNT:
num, curr, _, _ = cast(AmountTuple, token.val)
token.txt = (
cls.float(num, case="nf", gender=_currency_to_gender(curr))
+ " "
+ cls.currency(curr)
)

elif token.kind == TOK.EMAIL:
token.txt = cls.email(token.txt)

elif token.kind == TOK.DATEABS:
token.txt = cls.date(token.txt, case="þf")

elif token.kind == TOK.DATEREL:
token.txt = cls.date(token.txt, case="þf")

elif token.kind == TOK.TIMESTAMPABS:
token.txt = cls.time(cls.date(token.txt, case="þf"))

elif token.kind == TOK.TIMESTAMPREL:
token.txt = cls.time(cls.date(token.txt, case="þf"))

elif token.kind == TOK.MEASUREMENT:
unit, num = cast(MeasurementTuple, token.val)
pass # TODO
# TOK.MEASUREMENT: lambda tok, term: tok.txt, SI_UNITS in tokenizer

elif token.kind == TOK.NUMWLETTER:
num, letter = cast(PunctuationTuple, token.val)
token.txt = (
cls.number(num, case="nf", gender="hk") + " " + cls.spell(letter)
)

elif token.kind == TOK.DOMAIN:
token.txt = cls.domain(token.txt)

elif token.kind == TOK.HASHTAG:
token.txt = f"myllumerki {token.txt.lstrip('#')}"

elif token.kind == TOK.MOLECULE:
token.txt = cls.molecule(token.txt)

elif token.kind == TOK.SSN:
token.txt = cls.digits(token.txt)

elif token.kind == TOK.USERNAME:
token.txt = cls.username(token.txt)

elif token.kind == TOK.SERIALNUMBER:
token.txt = cls.digits(token.txt)

elif token.kind == TOK.COMPANY:
token.txt = cls.entity(token.txt)

return detokenize(tokens)

0 comments on commit a3a9055

Please sign in to comment.