speech-scoring/tts_samplegen.py

import objc
from AppKit import NSSpeechSynthesizer, NSSpeechInputModeProperty
from AppKit import NSSpeechModePhoneme
from Foundation import NSURL
import json
import random
import os
import re
import subprocess

OUTPUT_NAME = 'audio'
dest_dir = os.path.abspath('.') + '/outputs/' + OUTPUT_NAME + '/'
dest_file = './outputs/' + OUTPUT_NAME + '.csv'


def create_dir(direc):
    if not os.path.exists(direc):
        os.mkdir(direc)


def dest_filename(n, v, r, t):
    return '{}-{}-{}-{}-'.format(n, v, r,
                                 t) + str(random.randint(0, 10000)) + '.aiff'


def dest_path(v, r, n):
    return dest_dir + v + '/' + r + '/' + n


def cli_gen_audio(speech_cmd, rate, voice, out_path):
    subprocess.call(
        ['say', '-v', voice, '-r',
         str(rate), '-o', out_path, speech_cmd])


class SynthFile(object):
    """docstring for SynthFile."""

    def __init__(self, word, phon, filename, voice, rate, operation):
        super(SynthFile, self).__init__()
        self.word = word
        self.phoneme = phon
        self.filename = filename
        self.voice = voice
        self.rate = rate
        self.variant = operation

    def get_json(self):
        return {
            'filename': self.filename,
            'voice': self.voice,
            'rate': self.rate,
            'operation': self.operation
        }

    def get_csv(self):
        return '{},{},{},{},{}\n'.format(self.word, self.phoneme, self.voice,
                                         self.rate, self.variant,
                                         self.filename)


class SynthVariant(object):
    """docstring for SynthVariant."""

    def __init__(self, identifier, rate):
        super(SynthVariant, self).__init__()
        self.synth = NSSpeechSynthesizer.alloc().initWithVoice_(identifier)
        self.synth.setVolume_(100)
        self.synth.setRate_(rate)
        self.phone_synth = NSSpeechSynthesizer.alloc().initWithVoice_(
            identifier)
        self.phone_synth.setVolume_(100)
        self.phone_synth.setRate_(rate)
        self.phone_synth.setObject_forProperty_error_(
            NSSpeechModePhoneme, NSSpeechInputModeProperty, None)
        self.identifier = identifier
        self.rate = rate
        self.name = identifier.split('.')[-1]

    def __repr__(self):
        return 'Synthesizer[{} - {}]({})'.format(self.name, self.rate)

    def generate_audio(self, word, variant):
        orig_phon, phoneme, phon_cmd = self.synth.phonemesFromText_(
            word), '', word
        if variant == 'low':
            # self.synth.startSpeakingString_toURL_(word,d_url)
            phoneme = orig_phon
        elif variant == 'medium':
            phoneme = re.sub('[0-9]', '', orig_phon)
            phon_cmd = '[[inpt PHON]] ' + phoneme
        elif variant == 'high':
            phoneme = orig_phon
            phon_cmd = word
        # elif variant == 'long':
        # if phon != '':
        # self.phone_synth.startSpeakingString_toURL_(phon,d_url)
        # else:
        #     self.synth.startSpeakingString_toURL_(word,d_url)
        fname = dest_filename(word, phoneme, self.name, self.rate)
        d_path = dest_path(self.name, self.rate, fname)
        # d_url = NSURL.fileURLWithPath_(d_path)
        cli_gen_audio(phon_cmd, self.rate, self.name, d_path)
        return SynthFile(word, phoneme, fname, self.name, self.rate, variant)


def synth_generator():
    voices_installed = NSSpeechSynthesizer.availableVoices()
    voice_attrs = [
        NSSpeechSynthesizer.attributesForVoice_(v) for v in voices_installed
    ]
    us_voices_ids = [
        v['VoiceIdentifier'] for v in voice_attrs
        if v['VoiceLanguage'] == 'en-US'
        and v['VoiceIdentifier'].split('.')[-1][0].isupper()
    ]
    # us_voices_ids = ['com.apple.speech.synthesis.voice.Fred',
    #                  'com.apple.speech.synthesis.voice.Alex',
    #                  'com.apple.speech.synthesis.voice.Victoria']
    # voice_rates = list(range(150,221,(220-180)//4))
    voice_rates = [150, 180, 210, 250]
    voice_synths = []
    create_dir(dest_dir)
    for v in us_voices_ids:
        for r in voice_rates:
            create_dir(dest_dir + v + '/' + r)
            voice_synths.append(SynthVariant(v, r))

    def synth_for_words(words):
        all_synths = []
        for w in words:
            for s in voice_synths:
                for v in ['low', 'medium', 'high']:
                    all_synths.append(s.generate_audio(w, v))
        return all_synths

    return synth_for_words


def write_synths(synth_list, fname, csv=False):
    f = open(fname, 'w')
    if csv:
        for s in synth_list:
            f.write(s.get_csv())
    else:
        json.dump([s.get_json() for s in synth_list], f)
    f.close()


def generate_audio_for_stories():
    stories_data = json.load(open('./inputs/all_stories_hs.json'))
    word_list = [t[0] for i in stories_data.values() for t in i]
    words_audio_synth = synth_generator()
    return words_audio_synth(word_list)


# words_audio_synth = synth_generator()
# synth = NSSpeechSynthesizer.alloc().init()
# voices_installed = NSSpeechSynthesizer.availableVoices()
# voice_attrs = [NSSpeechSynthesizer.attributesForVoice_(v) for v in  voices_installed]
# us_voices_ids = [v['VoiceIdentifier'] for v in voice_attrs if v['VoiceLanguage'] == 'en-US' and v['VoiceIdentifier'].split('.')[-1][0].isupper()]
# synth.setVoice_(us_voices_ids[2])
# synth.startSpeakingString_('your')
# fname = dest_filename(word,self.name,self.rate,self.operation)
# d_path = dest_path(fname)
# d_url = dest_url(d_path)

synths = synth_generator()([OUTPUT_NAME])
# synths = generate_audio_for_stories()
write_synths(synths, dest_file, True)
# write_synths(synths,'./outputs/synths.json')