malar
/
speech-scoring
1
0
Fork 0
Code Issues Pull Requests Releases Wiki Activity

1. using cli say instead of api since api generates empty responses sometimes 

2. generating all words voices for each variants
master
Malar Kannan 2017-10-05 11:02:38 +05:30
parent 9d700f18ca
commit 0337f0d5be
1 changed files with 51 additions and 28 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

79
tts-wav-gen.py
View File

@ -1,16 +1,20 @@
import objc import objc
from AppKit import NSSpeechSynthesizer,NSSpeechInputModeProperty,NSSpeechModePhoneme from AppKit import NSSpeechSynthesizer,NSSpeechInputModeProperty,NSSpeechModePhoneme
from Foundation import NSURL,NSError from Foundation import NSURL,NSError,NSObject
import json import json
import random import random
import os import os
import re import re
import subprocess
dest_filename = lambda p: p+str(random.randint(0,10000))+'.aiff' dest_filename = lambda p: p+str(random.randint(0,10000))+'.aiff'
dest_path = lambda p: os.path.abspath('.')+'/outputs/audio/'+p dest_path = lambda p: os.path.abspath('.')+'/outputs/audio/'+p
dest_url = lambda p: NSURL.fileURLWithPath_(dest_path(p)) dest_url = lambda p: NSURL.fileURLWithPath_(dest_path(p))
def cli_gen_audio(word,rate,voice,out_path):
subprocess.call(['say','-v',voice,'-r',str(rate),'-o',out_path,word])
class SynthFile(object): class SynthFile(object):
"""docstring for SynthFile.""" """docstring for SynthFile."""
def __init__(self,word, filename,voice,rate,operation): def __init__(self,word, filename,voice,rate,operation):
@ -32,49 +36,68 @@ class SynthVariant(object):
"""docstring for SynthVariant.""" """docstring for SynthVariant."""
def __init__(self,identifier,rate,op): def __init__(self,identifier,rate,op):
super(SynthVariant, self).__init__() super(SynthVariant, self).__init__()
sp = NSSpeechSynthesizer.alloc().init() self.synth = NSSpeechSynthesizer.alloc().initWithVoice_(identifier)
sp.setVolume_(100) self.synth.setVolume_(100)
sp.setVoice_(identifier) # sp.setVoice_(identifier)
sp.setRate_(rate) self.synth.setRate_(rate)
self.synth = sp self.phone_synth = NSSpeechSynthesizer.alloc().initWithVoice_(identifier)
p_syn = NSSpeechSynthesizer.alloc().init() self.phone_synth.setVolume_(100)
p_syn.setVolume_(100) self.phone_synth.setRate_(rate)
p_syn.setVoice_(identifier) self.phone_synth.setObject_forProperty_error_(NSSpeechModePhoneme,NSSpeechInputModeProperty,None)
p_syn.setRate_(rate)
p_syn.setObject_forProperty_error_(NSSpeechModePhoneme,NSSpeechInputModeProperty,None)
self.phone_synth = p_syn
self.identifier = identifier self.identifier = identifier
self.rate = rate self.rate = rate
self.name = identifier.split('.')[-1] self.name = identifier.split('.')[-1]
self.operation = op self.operation = op
def __repr__(self):
return 'Synthesizer[{} - {}]({})'.format(self.name,self.rate,self.operation)
def synth_file(self,word): def generate_audio(self,word):
fname = dest_filename(word) fname = dest_filename(word)
d_path = dest_path(fname)
d_url = dest_url(fname) d_url = dest_url(fname)
started = False
if self.operation == 'normal': if self.operation == 'normal':
self.synth.startSpeakingString_toURL_(word,d_url) # self.synth.startSpeakingString_toURL_(word,d_url)
cli_gen_audio(word,self.rate,self.name,d_path)
else: else:
orig_phon = self.synth.phonemesFromText_(word) orig_phon = self.synth.phonemesFromText_(word)
phon = re.sub('[0-9]','',orig_phon) phon = '[[inpt PHON]] '+re.sub('[0-9]','',orig_phon)
self.phone_synth.startSpeakingString_toURL_(phon,d_url) cli_gen_audio(phon,self.rate,self.name,d_path)
# if phon != '':
# self.phone_synth.startSpeakingString_toURL_(phon,d_url)
# else:
# self.synth.startSpeakingString_toURL_(word,d_url)
return SynthFile(word,fname,self.name,self.rate,self.operation) return SynthFile(word,fname,self.name,self.rate,self.operation)
def synth_file(self,word):
# s = objc.selector(self.generate_audio,signature=b"@@:@")
# obj = NSObject.alloc().init()
# sf = obj.performSelectorOnMainThread_withObject_waitUntilDone_(s,word,True)
# return sf
return self.generate_audio(word)
def synth_generator(): def synth_generator():
voices_installed = NSSpeechSynthesizer.availableVoices() voices_installed = NSSpeechSynthesizer.availableVoices()
voice_attrs = [NSSpeechSynthesizer.attributesForVoice_(v) for v in voices_installed] 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'] 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_rates = list(range(180,221,(220-180)//4))
voice_synths = [] voice_synths = []
variants = ['normal','phoneme'] variants = ['normal','phoneme']
for v in us_voices_ids: for v in us_voices_ids:
for r in voice_rates: for r in voice_rates:
for o in variants: for o in variants:
voice_synths.append(SynthVariant(v,r,o)) voice_synths.append(SynthVariant(v,r,o))
def synth_for_word(word): def synth_for_words(words):
return [s.synth_file(word) for s in voice_synths] all_synths = []
return synth_for_word for s in voice_synths:
for w in words:
all_synths.append(s.synth_file(w))
# print(s)
# return [s.synth_file(word) for s in voice_synths]
return all_synths
return synth_for_words
def write_synths(synth_list,fname,csv=False): def write_synths(synth_list,fname,csv=False):
f = open(fname,'w') f = open(fname,'w')
@ -88,14 +111,14 @@ def write_synths(synth_list,fname,csv=False):
def generate_audio_for_stories(): def generate_audio_for_stories():
stories_data = json.load(open('./inputs/all_stories_hs.json')) 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_list = [t[0] for i in stories_data.values() for t in i]
word_audio_synth = synth_generator() words_audio_synth = synth_generator()
all_synths = [] # all_synths = []
for word in word_list: # for word in word_list[:1]:
words_synths = word_audio_synth(word) # words_synths = word_audio_synth(word)
all_synths.extend(words_synths) # all_synths.extend(words_synths)
return all_synths return words_audio_synth(word_list)
# synths = synth_generator()('education') # synths = synth_generator()(['education'])
synths = generate_audio_for_stories() synths = generate_audio_for_stories()
write_synths(synths,'./outputs/synth_data.csv',True) write_synths(synths,'./outputs/synth_data.csv',True)
write_synths(synths,'./outputs/synths.json') # write_synths(synths,'./outputs/synths.json')