coqui_textfile.py

from TTS.api import TTS
import pysbd
import os
import time
from datetime import timedelta

print(TTS.list_models())

model_name = "tts_models/multilingual/multi-dataset/your_tts"
sample_wav = "samples_of_voice_to_copy.wav"

# Init TTS
tts = TTS(model_name=model_name, gpu=True, progress_bar=True)

f = open('Part1.txt')

# "The Mysterious Island" from Gutenberg uses these weird characters that confuse it. There are better ways to do this.
fulltext = f.read(-1).replace("“", "\"").replace("”", "\"").replace("—", "-").replace("’", "'").replace("\" \"", "\". \"")

f.close()

# It crashes... I'll split it myself.
seg = pysbd.Segmenter(language="en", clean=True)
lines = seg.segment(fulltext)

start = time.time()
idx = 1
for line in lines:
    if len(line) < 5:
        continue

    # Text to speech to a file
    tts.tts_to_file(text=line, file_path="/tmp/output.wav", language="en", speaker_wav=sample_wav)
    if 1 == idx:
        os.system("ffmpeg -i /tmp/output.wav -y /tmp/output.aac")
    else:
        os.system("ffmpeg -i /tmp/output.wav -y /tmp/piece.aac")
        os.system("cat /tmp/piece.aac >> /tmp/output.aac; rm /tmp/piece.aac")

    elapsed = time.time() - start
    print("Processed line %i of %d [%s]" % (idx, len(lines), str(timedelta(seconds=elapsed))))

    idx = idx + 1

print("Done, I guess.")
os.system("mv /tmp/output.aac finished.aac")