speech-scoring/tts-wav-gen.py

102 lines
3.5 KiB
Python

import objc
from AppKit import NSSpeechSynthesizer,NSSpeechInputModeProperty,NSSpeechModePhoneme
from Foundation import NSURL,NSError
import json
import random
import os
import re
dest_filename = lambda p: p+str(random.randint(0,10000))+'.aiff'
dest_path = lambda p: os.path.abspath('.')+'/outputs/audio/'+p
dest_url = lambda p: NSURL.fileURLWithPath_(dest_path(p))
class SynthFile(object):
"""docstring for SynthFile."""
def __init__(self,word, filename,voice,rate,operation):
super(SynthFile, self).__init__()
self.word = word
self.filename = filename
self.voice = voice
self.rate = rate
self.operation = 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.voice,self.rate,self.operation,self.filename)
class SynthVariant(object):
"""docstring for SynthVariant."""
def __init__(self,identifier,rate,op):
super(SynthVariant, self).__init__()
sp = NSSpeechSynthesizer.alloc().init()
sp.setVolume_(100)
sp.setVoice_(identifier)
sp.setRate_(rate)
self.synth = sp
p_syn = NSSpeechSynthesizer.alloc().init()
p_syn.setVolume_(100)
p_syn.setVoice_(identifier)
p_syn.setRate_(rate)
p_syn.setObject_forProperty_error_(NSSpeechModePhoneme,NSSpeechInputModeProperty,None)
self.phone_synth = p_syn
self.identifier = identifier
self.rate = rate
self.name = identifier.split('.')[-1]
self.operation = op
def synth_file(self,word):
fname = dest_filename(word)
d_url = dest_url(fname)
if self.operation == 'normal':
self.synth.startSpeakingString_toURL_(word,d_url)
else:
orig_phon = self.synth.phonemesFromText_(word)
phon = re.sub('[0-9]','',orig_phon)
self.phone_synth.startSpeakingString_toURL_(phon,d_url)
return SynthFile(word,fname,self.name,self.rate,self.operation)
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']
voice_rates = list(range(180,221,(220-180)//5))
voice_synths = []
variants = ['normal','phoneme']
for v in us_voices_ids:
for r in voice_rates:
for o in variants:
voice_synths.append(SynthVariant(v,r,o))
def synth_for_word(word):
return [s.synth_file(word) for s in voice_synths]
return synth_for_word
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]
word_audio_synth = synth_generator()
all_synths = []
for word in word_list:
words_synths = word_audio_synth(word)
all_synths.extend(words_synths)
return all_synths
# synths = synth_generator()('education')
synths = generate_audio_for_stories()
write_synths(synths,'./outputs/synth_data.csv',True)
write_synths(synths,'./outputs/synths.json')