TTS Silero-v4

tts_silero_v4.py

import argparse
import sys

import num2words.base
import transliterate


def process_text(filename,tokens):
    f = open(filename,'r', encoding="utf-8")
    txt :str = f.read()
    f.close()
    txt = txt.lower()
    txt = txt.replace('…', '.')
    txt = txt.replace('...', '. ')
    txt = txt.replace('+', ' плюс ')
    txt = txt.replace('.\n', ' . ')
    txt = txt.replace('\n', ' . ')
    txt = txt.replace('\r', ' . ')
    txt = txt.replace('  ', ' ')

    #txt = txt.replace('.', '.\n')
    #lines = txt.split('\n')
    lines = txt.split(' ')

    # преобразовать цифры в текст
    # преобразовать латинские в русские
    newlines = []
    # множество латинских букв
    lat_lets = set(chr(i) for i in range (ord('A'),ord('Z')+1))
    lat_lets = lat_lets.union(set(chr(i) for i in range(ord('a'), ord('z') + 1)))
    for word in lines:
        if len(word) == 0:
            continue
        if any(map(str.isdigit, word)):
            text = []
            num = ''
            for c in word:
                if not c.isdigit():
                    if len(num)>0:
                        numtext = num2words.num2words(num,lang='ru')
                        text.append(numtext)
                        num = ''
                    text.append(c)
                else:
                    num = num + c
            if len(num) > 0:
                numtext = num2words.num2words(num, lang='ru')
                text.append(numtext)
            word = ''.join(text)

        if len(set(word).intersection(lat_lets)) > 0:
            text = []
            lets = ''
            for c in word:
                if not (c in lat_lets):
                    if len(lets) > 0:
                        letsrus = transliterate.translit(lets,'ru')
                        text.append(letsrus)
                        lets = ''
                    text.append(c)
                else:
                    lets = lets + c
            if len(lets) > 0:
                letsrus = transliterate.translit(lets, 'ru')
                text.append(letsrus)
            word = ''.join(text)

        newlines.append(word)
    lines = newlines
   # lines = [(x+'.') for x in lines]
    i=0
    newlines = []
    while i < len(lines):
        line = lines[i]

        if len(line) == 0:
            i = i + 1
            continue
        #if line[-1] != '.':
        #    line = line + "."

        newline = []
        for char in line:
            if char in tokens:
                newline.append(char)
            else:
                print("wrong char:",ord(char)," ",char)
                #newline.append(" ")

        # split too long text
        line = ''.join(newline)
        newlines.append(line)
        i = i + 1

    lines = newlines

    newlines = []
    i=0
    store = []
    while i < len(lines):
        sl = 0
        for a in store: #кастыль
            sl = sl + len(a)
        l = 0
        k = 0
        add_flag = False
        while True:
            if len(lines[i])+l+len(store)+sl < 900: # в том числе учесть добавляемые пробелы
                l = l + len(lines[i])+1
                store.append(lines[i])
                add_flag = True
                i = i + 1
                if i == len(lines):
                    break
            else:
                break
        sl = l
        last_dot = 0 # split my last dot

        for line in reversed(store):
            if line[-1]== '.':
                break
            last_dot += 1

        if not add_flag:
            addlines = store
            store = []
        else:
            if (last_dot != len(store)) and (last_dot != 0):
                addlines = store[:-last_dot]
                store = store[-last_dot:]
            else:
                addlines = store
                store = []


        newline = ' '.join(addlines)
        newlines.append(newline)

    lines = newlines
    return lines

import subprocess,io

def writeffmpeg(name,data):
    command = ['ffmpeg','-y','-f','f32le','-ar','16000','-ac','1','-i','-',str(name) + '.mp3']
    process = subprocess.Popen(command,stdin=subprocess.PIPE)
    process.stdin.write(data.tobytes())
    process.stdin.close()
    process.wait()

def writeffmpegrawlist(name, datalist, pathdir, pathfile):
    command = ['ffmpeg','-y','-f','f32le','-ar',str(sample_rate),'-ac','1','-i','-','-b:a','128k',pathdir+'\\'+pathfile+' '+str(name) + '.mp3']
    process = subprocess.Popen(command,stdin=subprocess.PIPE)
    for data in datalist:
        process.stdin.write(data.tobytes())
    process.stdin.close()
    process.wait()

def writeffmpegmp3(name,datalist):
    commandraw = ['ffmpeg', '-hide_banner', '-loglevel', 'fatal', '-y', '-i', '-',
               '-acodec', 'pcm_s16le', '-ac', '1', '-f', 's16le',  '-ar', '22050','-']

    commandmp3 = ['ffmpeg', '-y','-hide_banner', '-loglevel', 'fatal','-y',
                   '-f', 's16le', '-ar', '22050', '-ac', '1','-i','-',
                  str(name) + '.mp3']
    mp3bytes = io.BytesIO()
    rawbytes = io.BytesIO()
    processmp3 = subprocess.Popen(commandmp3, stdin=subprocess.PIPE)

    for data in datalist:
        processraw = subprocess.Popen(commandraw, stdin=subprocess.PIPE,
                                      stdout=subprocess.PIPE)
        rawdata,_ = processraw.communicate(data.getvalue())
        print('wait raw')
        processraw.wait()
        processmp3.stdin.write(rawdata)
    #rawbytes.close()
    processmp3.stdin.close()
    print('wait mp3')
    processmp3.wait()


import os
import torch

ars = "filename -h 48000 -s kseniya -b 1.txt".split(" ")
parser = argparse.ArgumentParser("Генератор аудиокниг (рус)",description="Генератор аудиокниг с использованием портабельной нейросети. Автор https://github.com/Filarius Сама нейросеть и большая благодарность https://github.com/snakers4/silero-models")
parser.add_argument("--threads",help="Количество потоков обработки (4)",default=4)
parser.add_argument("--gpu",help="Включить использование GPU (не тестировалось)",action='store_true',default=False)
parser.add_argument("--hz",help="Выбери Hz 8000 24000 48000",default=48000)
parser.add_argument("--speaker",help="Выбери диктора: aidar baya kseniya xenia",default='kseniya')
parser.add_argument("--book",help="Путь к книге в текстовом формате",required=False)
parser.add_argument("--output",help="Путь к папке для сохранения аудиокниги")
parser.add_argument("--accent",help="Включить авто-ударение",action='store_true',default=False)
parser.add_argument("--yo",help="Включить автоматическую простановку Ё",action='store_true',default=False)
#parser.add_argument("--bruteforce",help="Режим перебора голосов",action='store_true',default=False)

args = parser.parse_args()



#sample_rate = 48000
#speaker = 'xenia'
#put_accent=True
#put_yo=True

sample_rate = int(args.hz)
speaker = args.speaker
put_accent = args.accent
put_yo = args.yo
pathin = args.book
pathout = args.output if  (args.output) else "output"
if not os.path.isdir(pathout):
    os.makedirs(pathout)
threads = int(args.threads)
device = None
if args.gpu:    #DOTO
    device = torch.device('gpu')
else:
    device = torch.device('cpu')
torch.set_num_threads(threads)
local_file = 'model.data'

if not os.path.isfile(local_file):
    torch.hub.download_url_to_file('https://models.silero.ai/models/tts/ru/ru_v3.pt',
                                   local_file)

model = torch.package.PackageImporter(local_file).load_pickle("tts_models", "model")
model.to(device)
'''
if args.bruteforce:
    c = 1
    speaker = 'random'
    example_text = 'В недрах тундры выдры в г+етрах т+ырят в вёдра ядра к+едров.'
    while True:
        audio = model.apply_tts(text=example_text,
                                speaker=speaker,
                                sample_rate=sample_rate)
        audio = audio.numpy()
        writeffmpegrawlist(c + 10000, [audio], pathout, "random " + pathin)
        voice_path = pathout + "\\random " + str(c + 10000)+'.data'
        model.save_random_voice(voice_path)
        c += 1
'''


speakers = ['aidar', 'baya', 'kseniya', 'xenia', 'random']

tokens = model.symbols
lines = process_text(pathin, tokens)

audios = []
i = 0
counter = 1
while i < len(lines):
    print("part:",i+1," of ",len(lines)," length:",len(lines[i]))
    print(lines[i])

    audio = model.apply_tts(text=lines[i],
                            speaker=speaker,
                            sample_rate=sample_rate,
                            put_accent=put_accent,
                            put_yo=put_yo)
    audios.append(audio.numpy())
    if ((i+1)%50)==0:
        writeffmpegrawlist(counter+10000, audios, pathout, pathin)
        counter += 1
        audios = []
    i = i + 1

if ((i)%50)!=0:
    writeffmpegrawlist(counter + 10000, audios, pathout, pathin)



print("I'm DONE !")