WordReference pseudo-TTS

sample.sh

python wordreferencespeak.py \
  never gonna give you up \
  never gonna let you down \
  never gonna run around and desert you

wordreferencespeak.py

import functools
import lxml.html
import random
import requests
import subprocess
import tempfile
import contextlib


@functools.lru_cache(128)
def get_mp3(word):
    page = lxml.html.fromstring(requests.get('http://www.wordreference.com/enfr/{}'.format(word)).content)
    return [el.find('source').get('src') for el in page.xpath('//*[@id="listen_widget"]//audio')]


def get_urls(sentence):
    for word in sentence:
        accents = get_mp3(word)
        try:
            yield 'http://www.wordreference.com/' + random.choice(accents)
        except IndexError:
            raise ValueError("Unspeakable word: " + word)


if __name__ == '__main__':
    import sys

    postprocess = False
    sentence = ' '.join(sys.argv[1:]).lower().split()

    with contextlib.ExitStack() as stack:
        fout_names = []
        for url in get_urls(sentence):
            fin = stack.enter_context(tempfile.NamedTemporaryFile(suffix='.mp3'))
            fin.write(requests.get(url).content)
            fin.flush()
            if postprocess:
                fout = stack.enter_context(tempfile.NamedTemporaryFile(suffix='.mp3'))
                cmd = ['sox', fin.name, fout.name, 'silence', '1', '0.05', '0.03%', 'reverse', 'silence', '1', '0.1', '5%', 'reverse']
                subprocess.check_call(cmd)
                fout_names.append(fout.name)
            else:
                fout_names.append(fin.name)

        subprocess.check_call(['cvlc', '--quiet'] + fout_names)