iftaken
2 years ago
parent
5e714ecb4a
commit
a488ec8342
|
Before Width: | Height: | Size: 84 KiB After Width: | Height: | Size: 106 KiB |
@ -0,0 +1,12 @@
|
||||
###########################################################
|
||||
# PARAS SETTING #
|
||||
###########################################################
|
||||
# Set to -1 to indicate that the parameter is the same as the pretrained model configuration
|
||||
|
||||
batch_size: 10
|
||||
learning_rate: 0.0001 # learning rate
|
||||
num_snapshots: -1
|
||||
|
||||
# frozen_layers should be a list
|
||||
# if you don't need to freeze, set frozen_layers to []
|
||||
frozen_layers: ["encoder"]
|
||||
@ -1,13 +1,8 @@
|
||||
aiofiles
|
||||
faiss-cpu
|
||||
fastapi
|
||||
librosa
|
||||
numpy
|
||||
paddlenlp
|
||||
paddlepaddle
|
||||
paddlespeech
|
||||
praatio==5.0.0
|
||||
pydantic
|
||||
python-multipartscikit_learn
|
||||
SoundFile
|
||||
python-multipart
|
||||
scikit_learn
|
||||
starlette
|
||||
uvicorn
|
||||
|
||||
@ -0,0 +1,182 @@
|
||||
import os
|
||||
|
||||
from .util import run_cmd
|
||||
|
||||
|
||||
class SAT:
|
||||
def __init__(self):
|
||||
# pretrain model path
|
||||
self.zh_pretrain_model_path = os.path.realpath(
|
||||
"source/model/erniesat_aishell3_ckpt_1.2.0")
|
||||
self.en_pretrain_model_path = os.path.realpath(
|
||||
"source/model/erniesat_vctk_ckpt_1.2.0")
|
||||
self.cross_pretrain_model_path = os.path.realpath(
|
||||
"source/model/erniesat_aishell3_vctk_ckpt_1.2.0")
|
||||
|
||||
self.zh_voc_model_path = os.path.realpath(
|
||||
"source/model/hifigan_aishell3_ckpt_0.2.0")
|
||||
self.eb_voc_model_path = os.path.realpath(
|
||||
"source/model/hifigan_vctk_ckpt_0.2.0")
|
||||
self.cross_voc_model_path = os.path.realpath(
|
||||
"source/model/hifigan_aishell3_ckpt_0.2.0")
|
||||
|
||||
self.now_file_path = os.path.dirname(__file__)
|
||||
self.BIN_DIR = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../paddlespeech/t2s/exps/ernie_sat"))
|
||||
|
||||
def zh_synthesize_edit(self,
|
||||
old_str: str,
|
||||
new_str: str,
|
||||
input_name: os.PathLike,
|
||||
output_name: os.PathLike,
|
||||
task_name: str="synthesize",
|
||||
erniesat_ckpt_name: str="snapshot_iter_289500.pdz"):
|
||||
|
||||
if task_name not in ['synthesize', 'edit']:
|
||||
print("task name only in ['edit', 'synthesize']")
|
||||
return None
|
||||
|
||||
# 运行时的 PYTHONPATH
|
||||
PYTHONPATH = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../examples/aishell3/ernie_sat"))
|
||||
|
||||
# 推理文件配置
|
||||
config_path = os.path.join(self.zh_pretrain_model_path, "default.yaml")
|
||||
phones_dict = os.path.join(self.zh_pretrain_model_path,
|
||||
"phone_id_map.txt")
|
||||
erniesat_ckpt = os.path.join(self.zh_pretrain_model_path,
|
||||
erniesat_ckpt_name)
|
||||
erniesat_stat = os.path.join(self.zh_pretrain_model_path,
|
||||
"speech_stats.npy")
|
||||
|
||||
voc = "hifigan_aishell3"
|
||||
voc_config = os.path.join(self.zh_voc_model_path, "default.yaml")
|
||||
voc_ckpt = os.path.join(self.zh_voc_model_path,
|
||||
"snapshot_iter_2500000.pdz")
|
||||
voc_stat = os.path.join(self.zh_voc_model_path, "feats_stats.npy")
|
||||
|
||||
cmd = self.get_cmd(
|
||||
task_name,
|
||||
input_name,
|
||||
old_str,
|
||||
new_str,
|
||||
config_path,
|
||||
phones_dict,
|
||||
erniesat_ckpt,
|
||||
erniesat_stat,
|
||||
voc,
|
||||
voc_config,
|
||||
voc_ckpt,
|
||||
voc_stat,
|
||||
output_name,
|
||||
source_lang="zh",
|
||||
target_lang="zh")
|
||||
|
||||
return run_cmd(cmd, output_name)
|
||||
|
||||
def crossclone(self,
|
||||
old_str: str,
|
||||
new_str: str,
|
||||
input_name: os.PathLike,
|
||||
output_name: os.PathLike,
|
||||
source_lang: str,
|
||||
target_lang: str,
|
||||
erniesat_ckpt_name: str="snapshot_iter_489000.pdz"):
|
||||
PYTHONPATH = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../examples/aishell3_vctk/ernie_sat"))
|
||||
# 推理文件配置
|
||||
config_path = os.path.join(self.cross_pretrain_model_path,
|
||||
"default.yaml")
|
||||
phones_dict = os.path.join(self.cross_pretrain_model_path,
|
||||
"phone_id_map.txt")
|
||||
erniesat_ckpt = os.path.join(self.cross_pretrain_model_path,
|
||||
erniesat_ckpt_name)
|
||||
erniesat_stat = os.path.join(self.cross_pretrain_model_path,
|
||||
"speech_stats.npy")
|
||||
|
||||
voc = "hifigan_aishell3"
|
||||
voc_config = os.path.join(self.cross_voc_model_path, "default.yaml")
|
||||
voc_ckpt = os.path.join(self.cross_voc_model_path,
|
||||
"snapshot_iter_2500000.pdz")
|
||||
voc_stat = os.path.join(self.cross_voc_model_path, "feats_stats.npy")
|
||||
task_name = "synthesize"
|
||||
cmd = self.get_cmd(task_name, input_name, old_str, new_str, config_path,
|
||||
phones_dict, erniesat_ckpt, erniesat_stat, voc,
|
||||
voc_config, voc_ckpt, voc_stat, output_name,
|
||||
source_lang, target_lang)
|
||||
|
||||
return run_cmd(cmd, output_name)
|
||||
|
||||
def en_synthesize_edit(self,
|
||||
old_str: str,
|
||||
new_str: str,
|
||||
input_name: os.PathLike,
|
||||
output_name: os.PathLike,
|
||||
task_name: str="synthesize",
|
||||
erniesat_ckpt_name: str="snapshot_iter_199500.pdz"):
|
||||
PYTHONPATH = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../examples/vctk/ernie_sat"))
|
||||
|
||||
# 推理文件配置
|
||||
config_path = os.path.join(self.en_pretrain_model_path, "default.yaml")
|
||||
phones_dict = os.path.join(self.en_pretrain_model_path,
|
||||
"phone_id_map.txt")
|
||||
erniesat_ckpt = os.path.join(self.en_pretrain_model_path,
|
||||
erniesat_ckpt_name)
|
||||
erniesat_stat = os.path.join(self.en_pretrain_model_path,
|
||||
"speech_stats.npy")
|
||||
|
||||
voc = "hifigan_aishell3"
|
||||
voc_config = os.path.join(self.zh_voc_model_path, "default.yaml")
|
||||
voc_ckpt = os.path.join(self.zh_voc_model_path,
|
||||
"snapshot_iter_2500000.pdz")
|
||||
voc_stat = os.path.join(self.zh_voc_model_path, "feats_stats.npy")
|
||||
|
||||
cmd = self.get_cmd(
|
||||
task_name,
|
||||
input_name,
|
||||
old_str,
|
||||
new_str,
|
||||
config_path,
|
||||
phones_dict,
|
||||
erniesat_ckpt,
|
||||
erniesat_stat,
|
||||
voc,
|
||||
voc_config,
|
||||
voc_ckpt,
|
||||
voc_stat,
|
||||
output_name,
|
||||
source_lang="en",
|
||||
target_lang="en")
|
||||
|
||||
return run_cmd(cmd, output_name)
|
||||
|
||||
def get_cmd(self, task_name, input_name, old_str, new_str, config_path,
|
||||
phones_dict, erniesat_ckpt, erniesat_stat, voc, voc_config,
|
||||
voc_ckpt, voc_stat, output_name, source_lang, target_lang):
|
||||
cmd = f"""
|
||||
FLAGS_allocator_strategy=naive_best_fit \
|
||||
FLAGS_fraction_of_gpu_memory_to_use=0.01 \
|
||||
python3 {self.BIN_DIR}/synthesize_e2e.py \
|
||||
--task_name={task_name} \
|
||||
--wav_path={input_name} \
|
||||
--old_str='{old_str}' \
|
||||
--new_str='{new_str}' \
|
||||
--source_lang={source_lang} \
|
||||
--target_lang={target_lang} \
|
||||
--erniesat_config={config_path} \
|
||||
--phones_dict={phones_dict} \
|
||||
--erniesat_ckpt={erniesat_ckpt} \
|
||||
--erniesat_stat={erniesat_stat} \
|
||||
--voc={voc} \
|
||||
--voc_config={voc_config} \
|
||||
--voc_ckpt={voc_ckpt} \
|
||||
--voc_stat={voc_stat} \
|
||||
--output_name={output_name}
|
||||
"""
|
||||
|
||||
return cmd
|
||||
@ -0,0 +1,94 @@
|
||||
import os
|
||||
|
||||
from .util import run_cmd
|
||||
|
||||
|
||||
def find_max_ckpt(model_path):
|
||||
max_ckpt = 0
|
||||
for filename in os.listdir(model_path):
|
||||
if filename.endswith('.pdz'):
|
||||
files = filename[:-4]
|
||||
a1, a2, it = files.split("_")
|
||||
if int(it) > max_ckpt:
|
||||
max_ckpt = int(it)
|
||||
return max_ckpt
|
||||
|
||||
|
||||
class FineTune:
|
||||
def __init__(self):
|
||||
self.now_file_path = os.path.dirname(__file__)
|
||||
self.PYTHONPATH = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../examples/other/tts_finetune/tts3"))
|
||||
self.BIN_DIR = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../paddlespeech/t2s/exps/fastspeech2"))
|
||||
self.pretrained_model_dir = os.path.realpath(
|
||||
"source/model/fastspeech2_aishell3_ckpt_1.1.0")
|
||||
self.voc_model_dir = os.path.realpath(
|
||||
"source/model/hifigan_aishell3_ckpt_0.2.0")
|
||||
self.finetune_config = os.path.join("conf/tts3_finetune.yaml")
|
||||
|
||||
def finetune(self, input_dir, exp_dir='temp', epoch=100):
|
||||
mfa_dir = os.path.join(exp_dir, 'mfa_result')
|
||||
dump_dir = os.path.join(exp_dir, 'dump')
|
||||
output_dir = os.path.join(exp_dir, 'exp')
|
||||
lang = "zh"
|
||||
ngpu = 1
|
||||
|
||||
cmd = f"""
|
||||
python3 {self.PYTHONPATH}/finetune.py \
|
||||
--input_dir={input_dir} \
|
||||
--pretrained_model_dir={self.pretrained_model_dir} \
|
||||
--mfa_dir={mfa_dir} \
|
||||
--dump_dir={dump_dir} \
|
||||
--output_dir={output_dir} \
|
||||
--lang={lang} \
|
||||
--ngpu={ngpu} \
|
||||
--epoch={epoch} \
|
||||
--finetune_config={self.finetune_config}
|
||||
"""
|
||||
|
||||
print(cmd)
|
||||
|
||||
return run_cmd(cmd, exp_dir)
|
||||
|
||||
def synthesize(self, text, wav_name, out_wav_dir, exp_dir='temp'):
|
||||
|
||||
voc = "hifigan_aishell3"
|
||||
dump_dir = os.path.join(exp_dir, 'dump')
|
||||
output_dir = os.path.join(exp_dir, 'exp')
|
||||
text_path = os.path.join(exp_dir, 'sentences.txt')
|
||||
lang = "zh"
|
||||
ngpu = 1
|
||||
|
||||
model_path = f"{output_dir}/checkpoints"
|
||||
ckpt = find_max_ckpt(model_path)
|
||||
|
||||
# 生成对应的语句
|
||||
with open(text_path, "w", encoding='utf8') as f:
|
||||
f.write(wav_name + " " + text)
|
||||
|
||||
cmd = f"""
|
||||
FLAGS_allocator_strategy=naive_best_fit \
|
||||
FLAGS_fraction_of_gpu_memory_to_use=0.01 \
|
||||
python3 {self.BIN_DIR}/../synthesize_e2e.py \
|
||||
--am=fastspeech2_aishell3 \
|
||||
--am_config={self.pretrained_model_dir}/default.yaml \
|
||||
--am_ckpt={output_dir}/checkpoints/snapshot_iter_{ckpt}.pdz \
|
||||
--am_stat={self.pretrained_model_dir}/speech_stats.npy \
|
||||
--voc={voc} \
|
||||
--voc_config={self.voc_model_dir}/default.yaml \
|
||||
--voc_ckpt={self.voc_model_dir}/snapshot_iter_2500000.pdz \
|
||||
--voc_stat={self.voc_model_dir}/feats_stats.npy \
|
||||
--lang={lang} \
|
||||
--text={text_path} \
|
||||
--output_dir={out_wav_dir} \
|
||||
--phones_dict={dump_dir}/phone_id_map.txt \
|
||||
--speaker_dict={dump_dir}/speaker_id_map.txt \
|
||||
--spk_id=0
|
||||
"""
|
||||
|
||||
out_path = os.path.join(out_wav_dir, f"{wav_name}.wav")
|
||||
|
||||
return run_cmd(cmd, out_path)
|
||||
@ -0,0 +1,59 @@
|
||||
import os
|
||||
import shutil
|
||||
|
||||
from .util import run_cmd
|
||||
|
||||
|
||||
class VoiceCloneGE2E():
|
||||
def __init__(self):
|
||||
# Path 到指定路径上
|
||||
self.now_file_path = os.path.dirname(__file__)
|
||||
self.BIN_DIR = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../paddlespeech/t2s/exps"))
|
||||
# am
|
||||
self.am = "fastspeech2_aishell3"
|
||||
self.am_config = "source/model/fastspeech2_nosil_aishell3_vc1_ckpt_0.5/default.yaml"
|
||||
self.am_ckpt = "source/model/fastspeech2_nosil_aishell3_vc1_ckpt_0.5/snapshot_iter_96400.pdz"
|
||||
self.am_stat = "source/model/fastspeech2_nosil_aishell3_vc1_ckpt_0.5/speech_stats.npy"
|
||||
self.phones_dict = "source/model/fastspeech2_nosil_aishell3_vc1_ckpt_0.5/phone_id_map.txt"
|
||||
# voc
|
||||
self.voc = "pwgan_aishell3"
|
||||
self.voc_config = "source/model/pwg_aishell3_ckpt_0.5/default.yaml"
|
||||
self.voc_ckpt = "source/model/pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz"
|
||||
self.voc_stat = "source/model/pwg_aishell3_ckpt_0.5/feats_stats.npy"
|
||||
# ge2e
|
||||
self.ge2e_params_path = "source/model/ge2e_ckpt_0.3/step-3000000.pdparams"
|
||||
|
||||
def vc(self, text, input_wav, out_wav):
|
||||
|
||||
# input wav 需要形成临时单独文件夹
|
||||
_, full_file_name = os.path.split(input_wav)
|
||||
ref_audio_dir = os.path.realpath("tmp_dir/ge2e")
|
||||
if os.path.exists(ref_audio_dir):
|
||||
shutil.rmtree(ref_audio_dir)
|
||||
else:
|
||||
os.makedirs(ref_audio_dir, exist_ok=True)
|
||||
shutil.copy(input_wav, ref_audio_dir)
|
||||
|
||||
output_dir = os.path.dirname(out_wav)
|
||||
|
||||
cmd = f"""
|
||||
python3 {self.BIN_DIR}/voice_cloning.py \
|
||||
--am={self.am} \
|
||||
--am_config={self.am_config} \
|
||||
--am_ckpt={self.am_ckpt} \
|
||||
--am_stat={self.am_stat} \
|
||||
--voc={self.voc} \
|
||||
--voc_config={self.voc_config} \
|
||||
--voc_ckpt={self.voc_ckpt} \
|
||||
--voc_stat={self.voc_stat} \
|
||||
--ge2e_params_path={self.ge2e_params_path} \
|
||||
--text="{text}" \
|
||||
--input-dir={ref_audio_dir} \
|
||||
--output-dir={output_dir} \
|
||||
--phones-dict={self.phones_dict}
|
||||
"""
|
||||
|
||||
output_name = os.path.join(output_dir, full_file_name)
|
||||
return run_cmd(cmd, output_name=output_name)
|
||||
@ -0,0 +1,56 @@
|
||||
import os
|
||||
import shutil
|
||||
|
||||
from .util import run_cmd
|
||||
|
||||
|
||||
class VoiceCloneTDNN():
|
||||
def __init__(self):
|
||||
# Path 到指定路径上
|
||||
self.now_file_path = os.path.dirname(__file__)
|
||||
self.BIN_DIR = os.path.realpath(
|
||||
os.path.join(self.now_file_path,
|
||||
"../../../../paddlespeech/t2s/exps"))
|
||||
|
||||
self.am = "fastspeech2_aishell3"
|
||||
self.am_config = "source/model/fastspeech2_aishell3_ckpt_vc2_1.2.0/default.yaml"
|
||||
self.am_ckpt = "source/model/fastspeech2_aishell3_ckpt_vc2_1.2.0/snapshot_iter_96400.pdz"
|
||||
self.am_stat = "source/model/fastspeech2_aishell3_ckpt_vc2_1.2.0/speech_stats.npy"
|
||||
self.phones_dict = "source/model/fastspeech2_aishell3_ckpt_vc2_1.2.0/phone_id_map.txt"
|
||||
# voc
|
||||
self.voc = "pwgan_aishell3"
|
||||
self.voc_config = "source/model/pwg_aishell3_ckpt_0.5/default.yaml"
|
||||
self.voc_ckpt = "source/model/pwg_aishell3_ckpt_0.5/snapshot_iter_1000000.pdz"
|
||||
self.voc_stat = "source/model/pwg_aishell3_ckpt_0.5/feats_stats.npy"
|
||||
|
||||
def vc(self, text, input_wav, out_wav):
|
||||
# input wav 需要形成临时单独文件夹
|
||||
_, full_file_name = os.path.split(input_wav)
|
||||
ref_audio_dir = os.path.realpath("tmp_dir/tdnn")
|
||||
if os.path.exists(ref_audio_dir):
|
||||
shutil.rmtree(ref_audio_dir)
|
||||
else:
|
||||
os.makedirs(ref_audio_dir, exist_ok=True)
|
||||
shutil.copy(input_wav, ref_audio_dir)
|
||||
|
||||
output_dir = os.path.dirname(out_wav)
|
||||
|
||||
cmd = f"""
|
||||
python3 {self.BIN_DIR}/voice_cloning.py \
|
||||
--am={self.am} \
|
||||
--am_config={self.am_config} \
|
||||
--am_ckpt={self.am_ckpt} \
|
||||
--am_stat={self.am_stat} \
|
||||
--voc={self.voc} \
|
||||
--voc_config={self.voc_config} \
|
||||
--voc_ckpt={self.voc_ckpt} \
|
||||
--voc_stat={self.voc_stat} \
|
||||
--text="{text}" \
|
||||
--input-dir={ref_audio_dir} \
|
||||
--output-dir={output_dir} \
|
||||
--phones-dict={self.phones_dict} \
|
||||
--use_ecapa=True
|
||||
"""
|
||||
|
||||
output_name = os.path.join(output_dir, full_file_name)
|
||||
return run_cmd(cmd, output_name=output_name)
|
||||
@ -0,0 +1,88 @@
|
||||
import axios from 'axios'
|
||||
import {apiURL} from "./API.js"
|
||||
|
||||
// 上传音频-vc
|
||||
export async function vcUpload(params){
|
||||
const result = await axios.post(apiURL.VC_Upload, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 上传音频-sat
|
||||
export async function satUpload(params){
|
||||
const result = await axios.post(apiURL.SAT_Upload, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 上传音频-finetune
|
||||
export async function fineTuneUpload(params){
|
||||
const result = await axios.post(apiURL.FineTune_Upload, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 删除音频
|
||||
export async function vcDel(params){
|
||||
const result = await axios.post(apiURL.VC_Del, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 获取音频列表vc
|
||||
export async function vcList(){
|
||||
const result = await axios.get(apiURL.VC_List);
|
||||
return result
|
||||
}
|
||||
// 获取音频列表Sat
|
||||
export async function satList(){
|
||||
const result = await axios.get(apiURL.SAT_List);
|
||||
return result
|
||||
}
|
||||
|
||||
// 获取音频列表fineTune
|
||||
export async function fineTuneList(params){
|
||||
const result = await axios.post(apiURL.FineTune_List, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// fineTune 一键重置 获取新的文件夹
|
||||
export async function fineTuneNewDir(){
|
||||
const result = await axios.get(apiURL.FineTune_NewDir);
|
||||
return result
|
||||
}
|
||||
|
||||
// 获取音频数据
|
||||
export async function vcDownload(params){
|
||||
const result = await axios.post(apiURL.VC_Download, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 获取音频数据Base64
|
||||
export async function vcDownloadBase64(params){
|
||||
const result = await axios.post(apiURL.VC_Download_Base64, params);
|
||||
return result
|
||||
}
|
||||
|
||||
|
||||
// 克隆合成G2P
|
||||
export async function vcCloneG2P(params){
|
||||
const result = await axios.post(apiURL.VC_CloneG2p, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 克隆合成SAT
|
||||
export async function vcCloneSAT(params){
|
||||
const result = await axios.post(apiURL.VC_CloneSAT, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 克隆合成 - finetune 微调
|
||||
export async function vcCloneFineTune(params){
|
||||
const result = await axios.post(apiURL.VC_CloneFineTune, params);
|
||||
return result
|
||||
}
|
||||
|
||||
// 克隆合成 - finetune 合成
|
||||
export async function vcCloneFineTuneSyn(params){
|
||||
const result = await axios.post(apiURL.VC_CloneFineTuneSyn, params);
|
||||
return result
|
||||
}
|
||||
|
||||
|
||||
Loading…
Reference in new issue