leon-ai/leon
1
1
Fork 0
mirror of https://github.com/leon-ai/leon.git synced 2024-08-17 06:00:33 +03:00
Code Issues Projects Releases Wiki Activity

feat(python tcp server): force TTS BERT local files

Browse Source
This commit is contained in:
louistiti 2024-06-06 08:43:06 +08:00
parent 7c9487b62b
commit 4ce470d3e7
No known key found for this signature in database
GPG Key ID: 92CD6A2E497E1669
9 changed files with 52 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
.gitignore vendored
View File

@ -25,6 +25,9 @@ leon.json
bridges/python/src/Pipfile.lock
tcp_server/src/Pipfile.lock
tcp_server/src/lib/tts/models/*.pth
tcp_server/src/lib/tts/models/**/*.bin
tcp_server/src/lib/tts/models/**/*.json
tcp_server/src/lib/tts/models/**/*.txt
tcp_server/src/lib/asr/models/**/*.bin
!tcp_server/**/.gitkeep
!bridges/python/**/.gitkeep

19
scripts/setup/setup-python-dev-env.js
View File

@ -17,9 +17,9 @@ import {
PYTHON_TCP_SERVER_ASR_MODEL_GPU_HF_PREFIX_DOWNLOAD_URL,
PYTHON_TCP_SERVER_SRC_ASR_MODEL_PATH_FOR_GPU,
PYTHON_TCP_SERVER_SRC_ASR_MODEL_PATH_FOR_CPU,
PYTHON_TCP_SERVER_SRC_TTS_BERT_FRENCH_DIR_PATH,
// PYTHON_TCP_SERVER_SRC_TTS_BERT_FRENCH_DIR_PATH,
PYTHON_TCP_SERVER_SRC_TTS_BERT_BASE_DIR_PATH,
PYTHON_TCP_SERVER_TTS_BERT_FRENCH_MODEL_HF_PREFIX_DOWNLOAD_URL,
// PYTHON_TCP_SERVER_TTS_BERT_FRENCH_MODEL_HF_PREFIX_DOWNLOAD_URL,
PYTHON_TCP_SERVER_TTS_BERT_BASE_MODEL_HF_PREFIX_DOWNLOAD_URL
} from '@/constants'
import { CPUArchitectures, OSTypes } from '@/types'
@ -67,12 +67,12 @@ const ASR_CPU_MODEL_FILES = [
'tokenizer.json',
'vocabulary.txt'
]
const TTS_BERT_FRENCH_MODEL_FILES = [
'pytorch_model.bin',
/*const TTS_BERT_FRENCH_MODEL_FILES = [
'pytorch_model.bin', // Not needed? Compare with HF auto download in ~/.cache/huggingface/hub...
'config.json',
'vocab.txt',
'tokenizer_config.json'
]
]*/
const TTS_BERT_BASE_MODEL_FILES = [
'pytorch_model.bin',
'config.json',
@ -421,7 +421,7 @@ SPACY_MODELS.set('fr', {
process.exit(1)
}
}
const installTTSBERTFrenchModel = async () => {
/*const installTTSBERTFrenchModel = async () => {
try {
LogHelper.info('Installing TTS BERT French model...')
@ -448,7 +448,7 @@ SPACY_MODELS.set('fr', {
LogHelper.error(`Failed to install TTS BERT French model: ${e}`)
process.exit(1)
}
}
}*/
const installTTSBERTBaseModel = async () => {
try {
LogHelper.info('Installing TTS BERT base model...')
@ -520,7 +520,8 @@ SPACY_MODELS.set('fr', {
)
}
LogHelper.info(
// TODO: later when multiple languages are supported
/*LogHelper.info(
'Checking whether TTS BERT French language model files are downloaded...'
)
const areTTSBERTFrenchFilesDownloaded = fs.existsSync(
@ -536,7 +537,7 @@ SPACY_MODELS.set('fr', {
LogHelper.success(
'TTS BERT French language model files are already downloaded'
)
}
}*/
LogHelper.info('Checking whether the TTS model is installed...')
const isTTSModelInstalled = fs.existsSync(

2
tcp_server/src/lib/constants.py
View File

@ -18,6 +18,8 @@ TTS_MODEL_NAME = f'EN-Leon-{TTS_MODEL_VERSION}-G_{TTS_MODEL_ITERATION}'
TTS_MODEL_FILE_NAME = f'{TTS_MODEL_NAME}.pth'
TTS_LIB_PATH = os.path.join(LIB_PATH, 'tts')
TTS_MODEL_FOLDER_PATH = os.path.join(TTS_LIB_PATH, 'models')
TTS_BERT_FRENCH_MODEL_DIR_PATH = os.path.join(TTS_MODEL_FOLDER_PATH, 'bert-case-french-europeana-cased')
TTS_BERT_BASE_MODEL_DIR_PATH = os.path.join(TTS_MODEL_FOLDER_PATH, 'bert-base-uncased')
TTS_MODEL_CONFIG_PATH = os.path.join(TTS_MODEL_FOLDER_PATH, 'config.json')
TTS_MODEL_PATH = os.path.join(TTS_MODEL_FOLDER_PATH, TTS_MODEL_FILE_NAME)
IS_TTS_ENABLED = os.environ.get('LEON_TTS', 'true') == 'true'

5
tcp_server/src/lib/tts/text/__init__.py
View File

@ -22,9 +22,8 @@ def cleaned_text_to_sequence(cleaned_text, tones, language, symbol_to_id=None):
def get_bert(norm_text, word2ph, language, device):
from .english_bert import get_bert_feature as en_bert
from .french_bert import get_bert_feature as fr_bert
# from .french_bert import get_bert_feature as fr_bert
lang_bert_func_map = {"EN": en_bert,
'FR': fr_bert}
lang_bert_func_map = {"EN": en_bert}
bert = lang_bert_func_map[language](norm_text, word2ph, device)
return bert

7
tcp_server/src/lib/tts/text/cleaner.py
View File

@ -1,9 +1,10 @@
from . import english, french
from . import english
from . import cleaned_text_to_sequence
import copy
language_module_map = {"EN": english,
'FR': french}
# language_module_map = {"EN": english,
# 'FR': french}
language_module_map = {"EN": english}
def clean_text(text, language):

11
tcp_server/src/lib/tts/text/english.py
View File

@ -4,6 +4,9 @@ import re
from g2p_en import G2p
from transformers import AutoTokenizer
from lib.constants import TTS_BERT_BASE_MODEL_DIR_PATH
from . import symbols
from .english_utils.abbreviations import expand_abbreviations
@ -192,8 +195,12 @@ def text_normalize(text):
text = expand_abbreviations(text)
return text
model_id = 'bert-base-uncased'
tokenizer = AutoTokenizer.from_pretrained(model_id)
load_model_params = {
"pretrained_model_name_or_path": TTS_BERT_BASE_MODEL_DIR_PATH,
"local_files_only": True
}
tokenizer = AutoTokenizer.from_pretrained(**load_model_params)
def g2p_old(text):
tokenized = tokenizer.tokenize(text)
# import pdb; pdb.set_trace()

10
tcp_server/src/lib/tts/text/english_bert.py
View File

@ -2,9 +2,13 @@ import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM
import sys
from lib.constants import TTS_BERT_BASE_MODEL_DIR_PATH
model_id = 'bert-base-uncased'
tokenizer = AutoTokenizer.from_pretrained(model_id)
load_model_params = {
"pretrained_model_name_or_path": TTS_BERT_BASE_MODEL_DIR_PATH,
"local_files_only": True
}
tokenizer = AutoTokenizer.from_pretrained(**load_model_params)
model = None
def get_bert_feature(text, word2ph, device=None):
@ -18,7 +22,7 @@ def get_bert_feature(text, word2ph, device=None):
if not device:
device = "cuda"
if model is None:
model = AutoModelForMaskedLM.from_pretrained(model_id).to(
model = AutoModelForMaskedLM.from_pretrained(**load_model_params).to(
device
)
with torch.no_grad():

9
tcp_server/src/lib/tts/text/french.py
View File

@ -1,5 +1,7 @@
from transformers import AutoTokenizer
from lib.constants import TTS_BERT_FRENCH_MODEL_DIR_PATH
from .fr_phonemizer import cleaner as fr_cleaner
from .fr_phonemizer import fr_to_ipa
@ -16,8 +18,11 @@ def text_normalize(text):
text = fr_cleaner.french_cleaners(text)
return text
model_id = 'dbmdz/bert-base-french-europeana-cased'
tokenizer = AutoTokenizer.from_pretrained(model_id)
load_model_params = {
"pretrained_model_name_or_path": 'dbmdz/bert-base-french-europeana-cased',
"local_files_only": True
}
tokenizer = AutoTokenizer.from_pretrained(**load_model_params)
def g2p(text, pad_start_end=True, tokenized=None):
if tokenized is None:

11
tcp_server/src/lib/tts/text/french_bert.py
View File

@ -2,8 +2,13 @@ import torch
from transformers import AutoTokenizer, AutoModelForMaskedLM
import sys
model_id = 'dbmdz/bert-base-french-europeana-cased'
tokenizer = AutoTokenizer.from_pretrained(model_id)
from lib.constants import TTS_BERT_FRENCH_MODEL_DIR_PATH
load_model_params = {
"pretrained_model_name_or_path": TTS_BERT_FRENCH_MODEL_DIR_PATH,
"local_files_only": True
}
tokenizer = AutoTokenizer.from_pretrained(**load_model_params)
model = None
def get_bert_feature(text, word2ph, device=None):
@ -17,7 +22,7 @@ def get_bert_feature(text, word2ph, device=None):
if not device:
device = "cuda"
if model is None:
model = AutoModelForMaskedLM.from_pretrained(model_id).to(
model = AutoModelForMaskedLM.from_pretrained(**load_model_params).to(
device
)
with torch.no_grad():