Spaces:
Running
on
CPU Upgrade
Running
on
CPU Upgrade
skip web server
Browse files- app.py +18 -15
- resources/app/no_server.py +202 -0
app.py
CHANGED
|
@@ -7,6 +7,7 @@ from subprocess import Popen, PIPE
|
|
| 7 |
import threading
|
| 8 |
from huggingface_hub import hf_hub_download
|
| 9 |
import gradio as gr
|
|
|
|
| 10 |
|
| 11 |
hf_model_name = "Pendrokar/xvapitch_nvidia"
|
| 12 |
hf_cache_models_path = '/home/user/.cache/huggingface/hub/models--Pendrokar--xvapitch_nvidia/snapshots/61b10e60b22bc21c1e072f72f1108b9c2b21e94c/'
|
|
@@ -146,8 +147,9 @@ def load_model(voice_model_name):
|
|
| 146 |
|
| 147 |
print('Loading voice model...')
|
| 148 |
try:
|
| 149 |
-
|
| 150 |
-
|
|
|
|
| 151 |
current_voice_model = voice_model_name
|
| 152 |
|
| 153 |
with open(model_path + '.json', 'r', encoding='utf-8') as f:
|
|
@@ -211,9 +213,10 @@ def predict(
|
|
| 211 |
|
| 212 |
print('Synthesizing...')
|
| 213 |
try:
|
| 214 |
-
|
| 215 |
-
|
| 216 |
-
|
|
|
|
| 217 |
except requests.exceptions.RequestException as err:
|
| 218 |
print('FAILED to synthesize: {err}')
|
| 219 |
save_path = ''
|
|
@@ -487,18 +490,18 @@ with gr.Blocks(css=".arpabet {display: inline-block; background-color: gray; bor
|
|
| 487 |
if __name__ == "__main__":
|
| 488 |
# Run the web server in a separate thread
|
| 489 |
|
| 490 |
-
print('Attempting to connect to local xVASynth server...')
|
| 491 |
-
try:
|
| 492 |
-
|
| 493 |
-
|
| 494 |
-
except requests.exceptions.RequestException as err:
|
| 495 |
-
|
| 496 |
-
|
| 497 |
-
|
| 498 |
-
|
| 499 |
|
| 500 |
print('running Gradio interface')
|
| 501 |
demo.launch()
|
| 502 |
|
| 503 |
# Wait for the web server thread to finish (shouldn't be reached in normal execution)
|
| 504 |
-
web_server_thread.join()
|
|
|
|
| 7 |
import threading
|
| 8 |
from huggingface_hub import hf_hub_download
|
| 9 |
import gradio as gr
|
| 10 |
+
import resources.app.no_server as xvaserver
|
| 11 |
|
| 12 |
hf_model_name = "Pendrokar/xvapitch_nvidia"
|
| 13 |
hf_cache_models_path = '/home/user/.cache/huggingface/hub/models--Pendrokar--xvapitch_nvidia/snapshots/61b10e60b22bc21c1e072f72f1108b9c2b21e94c/'
|
|
|
|
| 147 |
|
| 148 |
print('Loading voice model...')
|
| 149 |
try:
|
| 150 |
+
xvaserver.loadModel(data)
|
| 151 |
+
# response = requests.post('http://0.0.0.0:8008/loadModel', json=data, timeout=60)
|
| 152 |
+
# response.raise_for_status() # If the response contains an HTTP error status code, raise an exception
|
| 153 |
current_voice_model = voice_model_name
|
| 154 |
|
| 155 |
with open(model_path + '.json', 'r', encoding='utf-8') as f:
|
|
|
|
| 213 |
|
| 214 |
print('Synthesizing...')
|
| 215 |
try:
|
| 216 |
+
xvaserver.synthesize(data)
|
| 217 |
+
# response = requests.post('http://0.0.0.0:8008/synthesize', json=data, timeout=60)
|
| 218 |
+
# response.raise_for_status() # If the response contains an HTTP error status code, raise an exception
|
| 219 |
+
# json_data = json.loads(response.text)
|
| 220 |
except requests.exceptions.RequestException as err:
|
| 221 |
print('FAILED to synthesize: {err}')
|
| 222 |
save_path = ''
|
|
|
|
| 490 |
if __name__ == "__main__":
|
| 491 |
# Run the web server in a separate thread
|
| 492 |
|
| 493 |
+
# print('Attempting to connect to local xVASynth server...')
|
| 494 |
+
# try:
|
| 495 |
+
# response = requests.get('http://0.0.0.0:8008')
|
| 496 |
+
# response.raise_for_status() # If the response contains an HTTP error status code, raise an exception
|
| 497 |
+
# except requests.exceptions.RequestException as err:
|
| 498 |
+
# print('Failed to connect to xVASynth!')
|
| 499 |
+
# web_server_thread = threading.Thread(target=run_xvaserver)
|
| 500 |
+
# print('Starting xVAServer thread')
|
| 501 |
+
# web_server_thread.start()
|
| 502 |
|
| 503 |
print('running Gradio interface')
|
| 504 |
demo.launch()
|
| 505 |
|
| 506 |
# Wait for the web server thread to finish (shouldn't be reached in normal execution)
|
| 507 |
+
# web_server_thread.join()
|
resources/app/no_server.py
ADDED
|
@@ -0,0 +1,202 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import os
|
| 2 |
+
import sys
|
| 3 |
+
import traceback
|
| 4 |
+
import multiprocessing
|
| 5 |
+
import json
|
| 6 |
+
|
| 7 |
+
torch_dml_device = None
|
| 8 |
+
|
| 9 |
+
multiprocessing.freeze_support()
|
| 10 |
+
|
| 11 |
+
# PROD = 'xVASynth.exe' in os.listdir(".")
|
| 12 |
+
PROD = True
|
| 13 |
+
sys.path.append("./resources/app")
|
| 14 |
+
|
| 15 |
+
# Saves me having to do backend re-compilations for every little UI hotfix
|
| 16 |
+
with open(f'{"./resources/app" if PROD else "."}/javascript/script.js', encoding="utf8") as f:
|
| 17 |
+
lines = f.read().split("\n")
|
| 18 |
+
APP_VERSION = lines[1].split('"v')[1].split('"')[0]
|
| 19 |
+
|
| 20 |
+
# Imports and logger setup
|
| 21 |
+
# ========================
|
| 22 |
+
try:
|
| 23 |
+
# import python.pyinstaller_imports
|
| 24 |
+
import numpy
|
| 25 |
+
|
| 26 |
+
import logging
|
| 27 |
+
from logging.handlers import RotatingFileHandler
|
| 28 |
+
import json
|
| 29 |
+
from socketserver import ThreadingMixIn
|
| 30 |
+
from python.audio_post import run_audio_post, prepare_input_audio, mp_ffmpeg_output, normalize_audio, start_microphone_recording, move_recorded_file
|
| 31 |
+
import ffmpeg
|
| 32 |
+
except:
|
| 33 |
+
print(traceback.format_exc())
|
| 34 |
+
with open("./DEBUG_err_imports.txt", "w+") as f:
|
| 35 |
+
f.write(traceback.format_exc())
|
| 36 |
+
|
| 37 |
+
# Pyinstaller hack
|
| 38 |
+
# ================
|
| 39 |
+
try:
|
| 40 |
+
def script_method(fn, _rcb=None):
|
| 41 |
+
return fn
|
| 42 |
+
def script(obj, optimize=True, _frames_up=0, _rcb=None):
|
| 43 |
+
return obj
|
| 44 |
+
import torch.jit
|
| 45 |
+
torch.jit.script_method = script_method
|
| 46 |
+
torch.jit.script = script
|
| 47 |
+
import torch
|
| 48 |
+
import tqdm
|
| 49 |
+
import regex
|
| 50 |
+
except:
|
| 51 |
+
with open("./DEBUG_err_import_torch.txt", "w+") as f:
|
| 52 |
+
f.write(traceback.format_exc())
|
| 53 |
+
# ================
|
| 54 |
+
# CPU_ONLY = not torch.cuda.is_available()
|
| 55 |
+
CPU_ONLY = True
|
| 56 |
+
|
| 57 |
+
try:
|
| 58 |
+
logger = logging.getLogger('serverLog')
|
| 59 |
+
logger.setLevel(logging.DEBUG)
|
| 60 |
+
server_log_path = f'{os.path.dirname(os.path.realpath(__file__))}/{"../../../" if PROD else ""}/server.log'
|
| 61 |
+
fh = RotatingFileHandler(server_log_path, maxBytes=2*1024*1024, backupCount=5)
|
| 62 |
+
fh.setLevel(logging.DEBUG)
|
| 63 |
+
ch = logging.StreamHandler()
|
| 64 |
+
ch.setLevel(logging.ERROR)
|
| 65 |
+
formatter = logging.Formatter('%(asctime)s - %(message)s')
|
| 66 |
+
fh.setFormatter(formatter)
|
| 67 |
+
ch.setFormatter(formatter)
|
| 68 |
+
logger.addHandler(fh)
|
| 69 |
+
logger.addHandler(ch)
|
| 70 |
+
logger.info(f'New session. Version: {APP_VERSION}. Installation: {"CPU" if CPU_ONLY else "CPU+GPU"} | Prod: {PROD} | Log path: {server_log_path}')
|
| 71 |
+
|
| 72 |
+
logger.orig_info = logger.info
|
| 73 |
+
|
| 74 |
+
def prefixed_log (msg):
|
| 75 |
+
logger.info(f'{logger.logging_prefix}{msg}')
|
| 76 |
+
|
| 77 |
+
|
| 78 |
+
def set_logger_prefix (prefix=""):
|
| 79 |
+
if len(prefix):
|
| 80 |
+
logger.logging_prefix = f'[{prefix}]: '
|
| 81 |
+
logger.log = prefixed_log
|
| 82 |
+
else:
|
| 83 |
+
logger.log = logger.orig_info
|
| 84 |
+
|
| 85 |
+
logger.set_logger_prefix = set_logger_prefix
|
| 86 |
+
logger.set_logger_prefix("")
|
| 87 |
+
|
| 88 |
+
except:
|
| 89 |
+
with open("./DEBUG_err_logger.txt", "w+") as f:
|
| 90 |
+
f.write(traceback.format_exc())
|
| 91 |
+
try:
|
| 92 |
+
logger.info(traceback.format_exc())
|
| 93 |
+
except:
|
| 94 |
+
pass
|
| 95 |
+
|
| 96 |
+
# if CPU_ONLY:
|
| 97 |
+
# torch_dml_device = torch.device("cpu")
|
| 98 |
+
|
| 99 |
+
|
| 100 |
+
try:
|
| 101 |
+
from python.plugins_manager import PluginManager
|
| 102 |
+
plugin_manager = PluginManager(APP_VERSION, PROD, CPU_ONLY, logger)
|
| 103 |
+
active_plugins = plugin_manager.get_active_plugins_count()
|
| 104 |
+
logger.info(f'Plugin manager loaded. {active_plugins} active plugins.')
|
| 105 |
+
except:
|
| 106 |
+
logger.info("Plugin manager FAILED.")
|
| 107 |
+
logger.info(traceback.format_exc())
|
| 108 |
+
|
| 109 |
+
plugin_manager.run_plugins(plist=plugin_manager.plugins["start"]["pre"], event="pre start", data=None)
|
| 110 |
+
|
| 111 |
+
|
| 112 |
+
# ======================== Models manager
|
| 113 |
+
# modelsPaths = {}
|
| 114 |
+
try:
|
| 115 |
+
from python.models_manager import ModelsManager
|
| 116 |
+
models_manager = ModelsManager(logger, PROD, device="cpu")
|
| 117 |
+
except:
|
| 118 |
+
logger.info("Models manager failed to initialize")
|
| 119 |
+
logger.info(traceback.format_exc())
|
| 120 |
+
# ========================
|
| 121 |
+
|
| 122 |
+
|
| 123 |
+
|
| 124 |
+
print("Models ready")
|
| 125 |
+
logger.info("Models ready")
|
| 126 |
+
|
| 127 |
+
|
| 128 |
+
global modelsPaths
|
| 129 |
+
post_data = ""
|
| 130 |
+
def loadModel(post_data):
|
| 131 |
+
logger.info("Direct: loadModel")
|
| 132 |
+
logger.info(post_data)
|
| 133 |
+
ckpt = post_data["model"]
|
| 134 |
+
modelType = post_data["modelType"]
|
| 135 |
+
instance_index = post_data["instance_index"] if "instance_index" in post_data else 0
|
| 136 |
+
modelType = modelType.lower().replace(".", "_").replace(" ", "")
|
| 137 |
+
post_data["pluginsContext"] = json.loads(post_data["pluginsContext"])
|
| 138 |
+
n_speakers = post_data["model_speakers"] if "model_speakers" in post_data else None
|
| 139 |
+
base_lang = post_data["base_lang"] if "base_lang" in post_data else None
|
| 140 |
+
|
| 141 |
+
|
| 142 |
+
plugin_manager.run_plugins(plist=plugin_manager.plugins["load-model"]["pre"], event="pre load-model", data=post_data)
|
| 143 |
+
models_manager.load_model(modelType, ckpt+".pt", instance_index=instance_index, n_speakers=n_speakers, base_lang=base_lang)
|
| 144 |
+
plugin_manager.run_plugins(plist=plugin_manager.plugins["load-model"]["post"], event="post load-model", data=post_data)
|
| 145 |
+
|
| 146 |
+
if modelType=="fastpitch1_1":
|
| 147 |
+
models_manager.models_bank["fastpitch1_1"][instance_index].init_arpabet_dicts()
|
| 148 |
+
|
| 149 |
+
return req_response
|
| 150 |
+
|
| 151 |
+
def synthesize(post_data):
|
| 152 |
+
logger.info("Direct: synthesize")
|
| 153 |
+
post_data["pluginsContext"] = json.loads(post_data["pluginsContext"])
|
| 154 |
+
instance_index = post_data["instance_index"] if "instance_index" in post_data else 0
|
| 155 |
+
|
| 156 |
+
|
| 157 |
+
# Handle the case where the vocoder remains selected on app start-up, with auto-HiFi turned off, but no setVocoder call is made before synth
|
| 158 |
+
continue_synth = True
|
| 159 |
+
if "waveglow" in post_data["vocoder"]:
|
| 160 |
+
waveglowPath = post_data["waveglowPath"]
|
| 161 |
+
req_response = models_manager.load_model(post_data["vocoder"], waveglowPath, instance_index=instance_index)
|
| 162 |
+
if req_response=="ENOENT":
|
| 163 |
+
continue_synth = False
|
| 164 |
+
|
| 165 |
+
device = post_data["device"] if "device" in post_data else models_manager.device_label
|
| 166 |
+
device = torch.device("cpu") if device=="cpu" else (torch_dml_device if CPU_ONLY else torch.device("cuda:0"))
|
| 167 |
+
models_manager.set_device(device, instance_index=instance_index)
|
| 168 |
+
|
| 169 |
+
if continue_synth:
|
| 170 |
+
plugin_manager.set_context(post_data["pluginsContext"])
|
| 171 |
+
plugin_manager.run_plugins(plist=plugin_manager.plugins["synth-line"]["pre"], event="pre synth-line", data=post_data)
|
| 172 |
+
|
| 173 |
+
modelType = post_data["modelType"]
|
| 174 |
+
text = post_data["sequence"]
|
| 175 |
+
pace = float(post_data["pace"])
|
| 176 |
+
out_path = post_data["outfile"]
|
| 177 |
+
base_lang = post_data["base_lang"] if "base_lang" in post_data else None
|
| 178 |
+
base_emb = post_data["base_emb"] if "base_emb" in post_data else None
|
| 179 |
+
pitch = post_data["pitch"] if "pitch" in post_data else None
|
| 180 |
+
energy = post_data["energy"] if "energy" in post_data else None
|
| 181 |
+
emAngry = post_data["emAngry"] if "emAngry" in post_data else None
|
| 182 |
+
emHappy = post_data["emHappy"] if "emHappy" in post_data else None
|
| 183 |
+
emSad = post_data["emSad"] if "emSad" in post_data else None
|
| 184 |
+
emSurprise = post_data["emSurprise"] if "emSurprise" in post_data else None
|
| 185 |
+
editorStyles = post_data["editorStyles"] if "editorStyles" in post_data else None
|
| 186 |
+
duration = post_data["duration"] if "duration" in post_data else None
|
| 187 |
+
speaker_i = post_data["speaker_i"] if "speaker_i" in post_data else None
|
| 188 |
+
useSR = post_data["useSR"] if "useSR" in post_data else None
|
| 189 |
+
useCleanup = post_data["useCleanup"] if "useCleanup" in post_data else None
|
| 190 |
+
vocoder = post_data["vocoder"]
|
| 191 |
+
globalAmplitudeModifier = float(post_data["globalAmplitudeModifier"]) if "globalAmplitudeModifier" in post_data else None
|
| 192 |
+
editor_data = [pitch, duration, energy, emAngry, emHappy, emSad, emSurprise, editorStyles]
|
| 193 |
+
old_sequence = post_data["old_sequence"] if "old_sequence" in post_data else None
|
| 194 |
+
|
| 195 |
+
model = models_manager.models(modelType.lower().replace(".", "_").replace(" ", ""), instance_index=instance_index)
|
| 196 |
+
req_response = model.infer(plugin_manager, text, out_path, vocoder=vocoder, \
|
| 197 |
+
speaker_i=speaker_i, editor_data=editor_data, pace=pace, old_sequence=old_sequence, \
|
| 198 |
+
globalAmplitudeModifier=globalAmplitudeModifier, base_lang=base_lang, base_emb=base_emb, useSR=useSR, useCleanup=useCleanup)
|
| 199 |
+
|
| 200 |
+
plugin_manager.run_plugins(plist=plugin_manager.plugins["synth-line"]["post"], event="post synth-line", data=post_data)
|
| 201 |
+
|
| 202 |
+
return req_response
|