text_to_speech.py

#!/usr/bin/python

# Converts text into spoken language saved to an mp3 file.


import argparse, base64, json, os, subprocess, sys, urllib


class PatchedArgumentParser(argparse.ArgumentParser):
    def error(self, message):
        sys.stderr.write('error: %s\n\n' % message)
        self.print_help()
        sys.exit(2)


sayVoiceByLang = {
    'de': 'Anna',
    'en': 'Samantha',
}
googleVoiceByLang = {
    'de': { 'languageCode': 'de-DE', 'name': 'de-DE-Wavenet-C' },
    'en': { 'languageCode': 'en-US', 'name': 'en-US-Wavenet-D' },
}
amazonVoiceByLang = {
    # See: https://docs.aws.amazon.com/de_de/polly/latest/dg/voicelist.html
    'de': 'Vicki',
    'en': 'Joanna',
}


textToSpeechDescription = """
The following text-to-speech engines are supported:
- With `--use-say` the text-to-speech engine of MacOS is used (command `say`).
- With `--use-amazon` Amazon Polly is used. Requires the AWS CLI to be installed and configured. See: https://aws.amazon.com/cli/
- With `--use-google-key=ABCD` Google text-to-speech is used. See: https://cloud.google.com/text-to-speech/

Amazon Polly sounds best, Google text-to-speech is second, MacOS `say` sounds worst.'
""".strip()

def addArgumentsToArgparser(argparser):
    argparser.add_argument('--lang', choices=['de', 'en'], default='de', help='The language (default: de)')
    argparser.add_argument('--use-say', action='store_true', default=None, help="If set, the MacOS tool `say` will be used.")
    argparser.add_argument('--use-amazon', action='store_true', default=None, help="If set, Amazon Polly is used. If missing the MacOS tool `say` will be used.")
    argparser.add_argument('--use-google-key', type=str, default=None, help="The API key of the Google text-to-speech account to use.")


def checkArgs(argparser, args):
    if not args.use_say and not args.use_amazon and args.use_google_key is None:
        print('ERROR: You have to provide one of the arguments `--use-say`, `--use-amazon` or `--use-google-key`\n')
        argparser.print_help()
        sys.exit(2)


def textToSpeechUsingArgs(text, targetFile, args):
    textToSpeech(text, targetFile, lang=args.lang, useAmazon=args.use_amazon, useGoogleKey=args.use_google_key)


def textToSpeech(text, targetFile, lang='de', useAmazon=False, useGoogleKey=None):
    print('\nGenerating: ' + targetFile + ' - ' + text)
    if useAmazon:
        response = subprocess.check_output(['aws', 'polly', 'synthesize-speech', '--output-format', 'mp3',
            '--voice-id', amazonVoiceByLang[lang], '--text-type', 'ssml',
            '--text', '<speak><amazon:effect name="drc"><prosody rate=\"+10%\">' + text + '</prosody></amazon:effect></speak>',
            targetFile])
    elif useGoogleKey:
        responseJson = postJson(
            'https://texttospeech.googleapis.com/v1beta1/text:synthesize?key=' + useGoogleKey,
            {
                'audioConfig': {
                    'audioEncoding': 'MP3',
                    'speakingRate': 1.0,
                    'pitch': 2.0,  # Default is 0.0
                    'sampleRateHertz': 44100,
                    'effectsProfileId': [ 'small-bluetooth-speaker-class-device' ]
                },
                'voice': googleVoiceByLang[lang],
                'input': { 'text': text }
            }
        )

        mp3Data = base64.b64decode(responseJson['audioContent'])

        with open(targetFile, 'wb') as f:
            f.write(mp3Data)
    else:
        subprocess.call([ 'say', '-v', sayVoiceByLang[lang], '-o', 'temp.aiff', text ])
        subprocess.call([ 'ffmpeg', '-y', '-i', 'temp.aiff', '-acodec', 'libmp3lame', '-ab', '128k', '-ac', '1', targetFile ])
        os.remove('temp.aiff')


def postJson(url, postBody, headers = None):
    cmd = ['curl']
    if headers is not None:
        for header in headers:
            cmd.extend(['-H', header])
    cmd.extend(['-H', 'Content-Type: application/json; charset=utf-8', '--data', json.dumps(postBody).encode('utf-8'), url])
    response = subprocess.check_output(cmd)
    return json.loads(response)


def postForm(url, formData):
    response = subprocess.check_output(['curl', '-H', 'Content-Type: application/x-www-form-urlencoded; charset=utf-8', '--data', urllib.urlencode(formData), url])
    return json.loads(response)


if __name__ == '__main__':
    argFormatter = lambda prog: argparse.RawDescriptionHelpFormatter(prog, max_help_position=30, width=100)
    argparser = PatchedArgumentParser(
        description=
            'Converts text into spoken language saved to an mp3 file.\n\n' +
            textToSpeechDescription,
        usage='%(prog)s -t "This is my text" -o my-output.mp3 [optional arguments...]',
        formatter_class=argFormatter)
    argparser.add_argument('-t', '--text', type=str, required=True, help='The text to convert into spoken language.')
    argparser.add_argument('-o', '--output', type=str, required=True, help='The output mp3 file to create')
    addArgumentsToArgparser(argparser)
    args = argparser.parse_args()


    checkArgs(argparser, args)

    if os.path.exists(args.output):
        print('ERROR: Output file alread exists: ' + os.path.abspath(args.output))
        sys.exit(1)


    textToSpeechUsingArgs(text=args.text, targetFile=args.output, args=args)