首页
学习
活动
专区
圈层
工具
发布
社区首页 >问答首页 >通过加载训练有素的模型进行多处理。(通过Python3)

问通过加载训练有素的模型进行多处理。(通过Python3)
EN

Stack Overflow用户
提问于 2018-06-06 14:27:47
回答 1查看 305关注 0票数 1

这就是我的情况:我有一个训练有素的语音合成模型。我将使用多处理来加快合成速度,即在每个CPU中预先加载模型,然后不断输入句子进行文本到语音的转换。

以下是我的尝试脚本:

代码语言:javascript
复制
####################################################################
#!/usr/bin/python3
####################################################################

from multiprocessing import Process, Pool, cpu_count
import os,time

####################################################################

from tacotron.demo_synthesizer import Synthesizer
from splitting_sent import splitting_para
import tensorflow as tf
from datasets import audio

####################################################################

from pypinyin import pinyin, Style

####################################################################

BASE_DIR = os.path.split(os.path.realpath(__file__))[0]
VOICE = BASE_DIR + "/tmp"
TXT = BASE_DIR + "/txt"
os.makedirs(VOICE, exist_ok=True)
os.makedirs(TXT, exist_ok=True)

####################################################################

def syn(py):
    synthesizer = Synthesizer()
    synthesizer.load("/path/to/the/model")
    wav_name = time.time()
    wav_path = VOICE + "/" + str(wav_name)
    wav = synthesizer.synthesize(py)
    audio.save_wav(wav, wav_path)

if __name__ =='__main__':

    with open(os.path.join(TXT, "content.txt"), "r") as f:
        lines = f.read().splitlines()
    lines = "".join(lines)

    sentences = splitting_para(lines)
    # splitting paragraph into individual sentences.

    py_list = []
    for sent in sentences:
        py_sent = pinyin(sent, style=Style.TONE3)
        py_sent = " ".join([i[0] for i in py_sent if i[0].isalnum()])
        py_list.append(py_sent)
    # as I am trying the Chinese TTS, it is the inevitable prerequisite step of translating Chinese character into Pinyin.

    print('Run the main process (%s).' % (os.getpid()))
    mainStart = time.time()
    p = Pool(cpu_count())
    for py in py_list:
        p.apply_async(syn,args=(py,))

    print('Waiting for all subprocesses done ...')
    p.close()
    p.join()
    print('All subprocesses done')
    mainEnd = time.time()
    print('All process ran %0.2f seconds.' % (mainEnd-mainStart))

我被这个问题卡住了:我只能将12个模型预加载到12个过程中来合成随机句子。然而,不可能在预加载模型中继续输入接下来的12个句子。这些过程被终止,随后完成了第一组12个句子的TTS。我完全迷失在这里了。:(

如果有任何建议帖子,我非常感谢。:)

EN

回答 1

Stack Overflow用户

发布于 2018-06-06 17:53:25

我从朋友那里得到了答案,答案如下:

代码语言:javascript
复制
####################################################################

import multiprocessing
from multiprocessing import Process, Pool, cpu_count, Queue
import os, time, sys

####################################################################

from tacotron.demo_synthesizer import Synthesizer
from splitting_sent import splitting_para
import tensorflow as tf
from datasets import audio

####################################################################

from pypinyin import pinyin, Style

####################################################################

BASE_DIR = os.path.split(os.path.realpath(__file__))[0]
VOICE = BASE_DIR + "/tmp"
TXT = BASE_DIR + "/txt"
os.makedirs(VOICE, exist_ok=True)
os.makedirs(TXT, exist_ok=True)

####################################################################

num_cpu = cpu_count() - 1
q = []
q_re = []

for tmp in range(num_cpu):
    q.append(Queue())

def syn(id):
    synthesizer = Synthesizer()
    synthesizer.load("/home/chris/Pictures/tts-server/logs-Tacotron/model.ckpt-1255000")
    print("LOADED at {} CPU".format(id))
    while not q[id].empty():
        input_py = q[id].get(True)
        print("Starting decode:", id)
        wav_name = time.time()
        wav_path = VOICE + "/" + str(wav_name)
        wav = synthesizer.synthesize(input_py)
        audio.save_wav(wav, wav_path)
        print("Decoded:", id)

if __name__ =='__main__':

    with open(os.path.join(TXT, "content.txt"), "r") as f:
        lines = f.read().splitlines()
    lines = "".join(lines)

    sentences = splitting_para(lines)

    py_list = []
    for sent in sentences:
        py_sent = pinyin(sent, style=Style.TONE3)
        py_sent = " ".join([i[0] for i in py_sent if i[0].isalnum()])
        py_list.append(py_sent)

    for x in range(num_cpu):
        p = multiprocessing.Process(target=syn, args=(x,))
        p.start()

    # decoding
    for index, py in enumerate(py_list):
        q[index % num_cpu].put(py)
票数 1
EN
页面原文内容由Stack Overflow提供。腾讯云小微IT领域专用引擎提供翻译支持
原文链接:

https://stackoverflow.com/questions/50713528

复制
相关文章

相似问题

领券
问题归档专栏文章快讯文章归档关键词归档开发者手册归档开发者手册 Section 归档