ONNX
piper-voices / _script /voicefest.py
synesthesiam's picture
Welsh Language Medium (#16)
5512791 verified
raw
history blame
12.3 kB
#!/usr/bin/env python3
import hashlib
import json
import unittest
from dataclasses import dataclass
from pathlib import Path
from typing import Set
_DIR = Path(__file__).parent
_REPO_DIR = _DIR.parent
@dataclass
class Language:
native: str
english: str
country: str
_LANGUAGES = {
"ar_JO": Language("العربية", "Arabic", "Jordan"),
"ca_ES": Language("Català", "Catalan", "Spain"),
"cs_CZ": Language("Čeština", "Czech", "Czech Republic"),
"cy_GB": Language("Cymraeg", "Welsh", "Great Britain"),
"da_DK": Language("Dansk", "Danish", "Denmark"),
"de_DE": Language("Deutsch", "German", "Germany"),
"el_GR": Language("Ελληνικά", "Greek", "Greece"),
"en_GB": Language("English", "English", "Great Britain"),
"en_US": Language("English", "English", "United States"),
"es_ES": Language("Español", "Spanish", "Spain"),
"es_MX": Language("Español", "Spanish", "Mexico"),
"fa_IR": Language("فارسی", "Farsi", "Iran"),
"fi_FI": Language("Suomi", "Finnish", "Finland"),
"fr_FR": Language("Français", "French", "France"),
"is_IS": Language("íslenska", "Icelandic", "Iceland"),
"it_IT": Language("Italiano", "Italian", "Italy"),
"hu_HU": Language("Magyar", "Hungarian", "Hungary"),
"ka_GE": Language("ქართული ენა", "Georgian", "Georgia"),
"kk_KZ": Language("қазақша", "Kazakh", "Kazakhstan"),
"lb_LU": Language("Lëtzebuergesch", "Luxembourgish", "Luxembourg"),
"ne_NP": Language("नेपाली", "Nepali", "Nepal"),
"nl_BE": Language("Nederlands", "Dutch", "Belgium"),
"nl_NL": Language("Nederlands", "Dutch", "Netherlands"),
"no_NO": Language("Norsk", "Norwegian", "Norway"),
"pl_PL": Language("Polski", "Polish", "Poland"),
"pt_BR": Language("Português", "Portuguese", "Brazil"),
"pt_PT": Language("Português", "Portuguese", "Portugal"),
"ro_RO": Language("Română", "Romanian", "Romania"),
"ru_RU": Language("Русский", "Russian", "Russia"),
"sk_SK": Language("Slovenčina", "Slovak", "Slovakia"),
"sl_SI": Language("Slovenščina", "Slovenian", "Slovenia"),
"sr_RS": Language("srpski", "Serbian", "Serbia"),
"sv_SE": Language("Svenska", "Swedish", "Sweden"),
"sw_CD": Language("Kiswahili", "Swahili", "Democratic Republic of the Congo"),
"tr_TR": Language("Türkçe", "Turkish", "Turkey"),
"uk_UA": Language("украї́нська мо́ва", "Ukrainian", "Ukraine"),
"vi_VN": Language("Tiếng Việt", "Vietnamese", "Vietnam"),
"zh_CN": Language("简体中文", "Chinese", "China"),
}
# -----------------------------------------------------------------------------
def add_languages():
for onnx_path in _REPO_DIR.rglob("*.onnx"):
config_path = f"{onnx_path}.json"
with open(config_path, "r", encoding="utf-8") as config_file:
config = json.load(config_file)
lang_code, dataset, quality = onnx_path.stem.split("-")
is_changed = False
lang_info = _LANGUAGES.get(lang_code)
assert lang_info is not None, f"Missing name for language: {lang_code}"
lang_family, lang_region = lang_code.split("_", maxsplit=1)
lang_dict = {
"code": lang_code,
"family": lang_family,
"region": lang_region,
"name_native": lang_info.native,
"name_english": lang_info.english,
"country_english": lang_info.country,
}
if "language" not in config:
config["language"] = lang_dict
is_changed = True
else:
current_lang_dict = config["language"]
if "code" not in current_lang_dict:
current_lang_dict["code"] = lang_dict["code"]
is_changed = True
if "family" not in current_lang_dict:
current_lang_dict["family"] = lang_dict["family"]
is_changed = True
if "region" not in current_lang_dict:
current_lang_dict["region"] = lang_dict["region"]
is_changed = True
if "name_native" not in current_lang_dict:
current_lang_dict["name_native"] = lang_dict["name_native"]
is_changed = True
if "name_english" not in current_lang_dict:
current_lang_dict["name_english"] = lang_dict["name_english"]
is_changed = True
if "country_english" not in current_lang_dict:
current_lang_dict["country_english"] = lang_dict["country_english"]
is_changed = True
if "dataset" not in config:
config["dataset"] = dataset
is_changed = True
if "quality" not in config["audio"]:
config["audio"]["quality"] = quality
is_changed = True
if is_changed:
with open(config_path, "w", encoding="utf-8") as config_file:
json.dump(config, config_file, ensure_ascii=False, indent=2)
# -----------------------------------------------------------------------------
class VoiceTest(unittest.TestCase):
def test_voices(self):
used_aliases: Set[str] = set()
for onnx_path in _REPO_DIR.rglob("*.onnx"):
with self.subTest(onnx_path=onnx_path):
self.assertGreater(onnx_path.stat().st_size, 0, "Empty onnx file")
# Load JSON config for voice
config_path = onnx_path.parent / f"{onnx_path.name}.json"
with open(config_path, "r", encoding="utf-8") as config_file:
config = json.load(config_file)
# Verify config
self.assertIn(
"piper_version", config, "Missing piper_version in config"
)
self.assertIn("language", config, "Missing language in config")
self.assertIn("dataset", config, "Missing dataset in config")
self.assertIn(
"quality", config["audio"], "Missing audio quality in config"
)
# Verify directory structure
# <lang_family>/<lang_code>/<dataset>/<quality>/
quality_dir = onnx_path.parent
dataset_dir = quality_dir.parent
lang_code_dir = dataset_dir.parent
lang_family_dir = lang_code_dir.parent
self.assertEqual(
lang_family_dir.name,
config["language"]["family"],
"Wrong lang family dir",
)
self.assertEqual(
lang_code_dir.name,
config["language"]["code"],
"Wrong lang code dir",
)
self.assertEqual(
dataset_dir.name, config["dataset"], "Wrong dataset dir"
)
self.assertEqual(
quality_dir.name, config["audio"]["quality"], "Wrong quality dir"
)
self.assertIn(lang_code_dir.name, _LANGUAGES, "Unknown language code")
# Verify file names
file_lang_code, file_dataset, file_quality = onnx_path.stem.split("-")
file_lang_family = file_lang_code.split("_", maxsplit=1)[0]
self.assertEqual(
file_lang_family,
config["language"]["family"],
"Wrong lang family file",
)
self.assertEqual(
file_lang_code, config["language"]["code"], "Wrong lang code file"
)
self.assertEqual(file_dataset, config["dataset"], "Wrong dataset file")
self.assertEqual(
file_quality, config["audio"]["quality"], "Wrong quality"
)
# Verify aliases are unique
aliases_path = onnx_path.parent / "ALIASES"
if aliases_path.exists():
with open(aliases_path, "r", encoding="utf-8") as aliases_file:
for alias in aliases_file:
alias = alias.strip()
self.assertNotIn(
alias,
used_aliases,
"Alias is already in use by another voice",
)
used_aliases.add(alias)
def run_tests() -> None:
runner = unittest.TextTestRunner()
result = runner.run(unittest.makeSuite(VoiceTest))
assert not result.failures, "Test failures"
# -----------------------------------------------------------------------------
def write_voices_json() -> None:
# {
# "<family>_<region>-<dataset>-<quality>": {
# "key": "<voice_key>",
# "name": "<dataset>",
# "language": {
# "code": "<family>_<region>",
# "family": "<family>",
# "region": "<region>",
# "name_native": "<native>",
# "name_english": "<english>",
# "country_english": "<country>",
# },
# "quality": "<quality>", // x_low, low, medium, high
# "num_speakers": int,
# "speaker_id_map": {
# "name": int,
# ...
# }
# "files": {
# "relative/path/to/file": {
# "size_bytes": int,
# "md5_digest": str, // hex
# },
# ...
# },
# "aliases": ["alias", ...],
# },
# ...
# }
voices = {}
for onnx_path in sorted(_REPO_DIR.rglob("*.onnx")):
voice_dir = onnx_path.parent
config_path = voice_dir / f"{onnx_path.name}.json"
with open(config_path, "r", encoding="utf-8") as config_file:
config = json.load(config_file)
quality = config["audio"]["quality"]
dataset = config["dataset"]
lang_code = config["language"]["code"]
lang_family, lang_region = lang_code.split("_", maxsplit=1)
lang_names = _LANGUAGES[lang_code]
voice_key = f"{lang_code}-{dataset}-{quality}"
model_card_path = voice_dir / "MODEL_CARD"
assert model_card_path.exists(), f"Missing {model_card_path}"
aliases: Set[str] = set()
aliases_path = voice_dir / "ALIASES"
if aliases_path.exists():
with open(aliases_path, "r", encoding="utf-8") as aliases_file:
for alias in aliases_file:
aliases.add(alias.strip())
voices[voice_key] = {
"key": voice_key,
"name": dataset,
"language": {
"code": lang_code,
"family": lang_family,
"region": lang_region,
"name_native": lang_names.native,
"name_english": lang_names.english,
"country_english": lang_names.country,
},
"quality": quality,
"num_speakers": config["num_speakers"],
"speaker_id_map": config.get("speaker_id_map", {}),
"files": {
str(file_path.relative_to(_REPO_DIR)): {
"size_bytes": file_path.stat().st_size,
"md5_digest": get_file_hash(file_path),
}
for file_path in (
onnx_path,
config_path,
model_card_path,
)
},
"aliases": sorted(list(aliases)),
}
with open(_REPO_DIR / "voices.json", "w", encoding="utf-8") as voices_file:
json.dump(voices, voices_file, indent=4, ensure_ascii=False)
def get_file_hash(path, bytes_per_chunk: int = 8192) -> str:
"""Hash a file in chunks using md5."""
path_hash = hashlib.md5()
with open(path, "rb") as path_file:
chunk = path_file.read(bytes_per_chunk)
while chunk:
path_hash.update(chunk)
chunk = path_file.read(bytes_per_chunk)
return path_hash.hexdigest()
# -----------------------------------------------------------------------------
if __name__ == "__main__":
add_languages()
run_tests()
print("Writing voices.json")
write_voices_json()