def _concatunits(self, utt, args):
""" Concatenates units and produces waveform via residual
excited LPC synthesis filter...
"""
unit_rel = utt.get_relation("Unit")
#concat:
unit_item = unit_rel.head_item
lpctrack = copy.deepcopy(
unit_item["selected_unit"]["candidate"]["lpc-coefs"])
residuals = window_residual(
unit_item["selected_unit"]["candidate"]["lpc-coefs"],
unit_item["selected_unit"]["candidate"]["residuals"])
unit_item = unit_item.next_item
while unit_item is not None:
temptrack = unit_item["selected_unit"]["candidate"]["lpc-coefs"]
#append lpccoefs to lpctrack:
lpctrack.times = np.concatenate(
(lpctrack.times, (temptrack.times + lpctrack.times[-1])))
lpctrack.values = np.concatenate(
(lpctrack.values, temptrack.values))
#append windowed residuals:
residuals.extend(
window_residual(
temptrack,
unit_item["selected_unit"]["candidate"]["residuals"]))
unit_item = unit_item.next_item
#overlap add residual:
lastsample = int(round(lpctrack.times[-1] * SAMPLERATE)) + int(
round(len(residuals[-1]) / 2))
residual = np.zeros(lastsample + 1)
for i, time in enumerate(lpctrack.times):
centersample = int(round(time * SAMPLERATE))
firstsample = centersample - int(len(residuals[i]) / 2)
residual[firstsample:firstsample + len(residuals[i])] += np.array(
residuals[i])
#synth filter:
samples = synth_filter(lpctrack.times, lpctrack.values,
residual.astype(np.float), SAMPLERATE)
#save in utterance:
w = Waveform()
w.samplerate = SAMPLERATE
w.samples = samples.astype("int16") #16bit samples
w.channels = 1
utt["waveform"] = w
return utt
def test(testfn="samples/data_001d.wav"):
aud = Waveform(testfn)
samples = zero_startends(aud.samples)
hh = make_hpf()
# mfreqz(h)
# show()
hpfsamples = filt(hh, samples)
detsamples = detectsimple(hpfsamples)
plot(samples)
plot(np.abs(hpfsamples))
plot(detsamples / 1000.0)
print(find_segs(detsamples))
show()
def get_wav(self):
assert self.donesynth
waveform = Waveform()
waveform.samples = np.zeros(LIBHTS.HTS_Engine_get_nsamples(
self.engine), np.int16) #16-bit samples
waveform.samplerate = int(
LIBHTS.HTS_Engine_get_sampling_frequency(self.engine))
waveform.channels = 1
for i in range(len(waveform.samples)):
waveform.samples[i] = LIBHTS.HTS_Engine_get_generated_speech(
self.engine, i) #copy
return waveform
def from_textgrid(voice):
""" Create aligned Utterances by synthesising to Segment level
from the orthography and simply copying label end times into
segment items as "end" feature.
"""
#Setup and create necessary dirs...
CWD = os.getcwd()
wav_dir = os.path.join(CWD, WAV_DIR)
uttwav_dir = os.path.join(CWD, UTTWAV_DIR)
transcr_location = os.path.join(CWD, ETC_DIR, TRANSCR_FILE)
textgrid_dir = os.path.join(CWD, TEXTGRID_DIR)
aligned_utts_dir = os.path.join(CWD, ALIGNED_UTT_DIR)
os.makedirs(aligned_utts_dir)
#update utts from textgrids...
transcriptions = load_transcriptions_schemefile(transcr_location)
alignments = sl.Corpus(textgrid_dir)
#################
for sc_utt, uttname, wavfilename in zip(
alignments.utterances, sorted(transcriptions),
sorted(glob(os.path.join(uttwav_dir, "*")))):
assert sc_utt.name == uttname, "Utterance missmatch..."
assert os.path.basename(wavfilename).startswith(
uttname), "Utterance missmatch..."
print("Synthesizing:", uttname)
utt = voice.synthesize(transcriptions[uttname], 'text-to-segments')
utt["file_id"] = uttname
utt = transplant_segtime_info(voice, sc_utt, utt)
#add waveform to utt:
utt["waveform"] = Waveform(wavfilename)
#save utt...
ttslab.tofile(
utt, os.path.join(aligned_utts_dir, ".".join([uttname, UTT_EXT])))
def make_aligned_utts(voice, transcriptions, sc_corpus, wav_dir, output_dir):
""" Make Word level utts and complete from 3-tier TextGrids...
"""
def copyuttfeats(u, u2):
for relname in ["Word", "Syllable"]:
items = u.gr(relname).as_list()
items2 = u2.gr(relname).as_list()
assert [i["name"] for i in items] == [i2["name"] for i2 in items2]
for i, i2 in zip(items, items2):
for k in i2:
if not k in i:
i[k] = i2[k]
return u
for sc_utt, uttname, wavfilename in zip(
sc_corpus.utterances, sorted(transcriptions),
sorted(glob(os.path.join(wav_dir, "*")))):
assert sc_utt.name == uttname, "Utterance missmatch..."
assert os.path.basename(wavfilename).startswith(
uttname), "Utterance missmatch..."
print("Synthesizing:", uttname)
utt = voice.synthesize(transcriptions[uttname], 'text-to-words')
utt["file_id"] = uttname
utt = complete_utt_from_textgrid(voice, sc_utt, utt)
utt2 = voice.synthesize(transcriptions[uttname], 'text-to-segments')
try:
utt = copyuttfeats(utt, utt2)
except AssertionError:
print("WARNING: could not copy item feats for %s" % utt["file_id"])
#add waveform to utt:
utt["waveform"] = Waveform(wavfilename)
#save utt...
ttslab.tofile(utt,
os.path.join(output_dir, ".".join([uttname, UTT_EXT])))
def make_aligned_utts(voice, transcriptions, sc_corpus, wav_dir, output_dir):
""" Make Word level utts and complete from 3-tier TextGrids...
"""
for sc_utt, uttname, wavfilename in zip(
sc_corpus.utterances, sorted(transcriptions),
sorted(glob(os.path.join(wav_dir, "*")))):
assert sc_utt.name == uttname, "Utterance missmatch..."
assert os.path.basename(wavfilename).startswith(
uttname), "Utterance missmatch..."
print("Synthesizing:", uttname)
utt = voice.synthesize(transcriptions[uttname], 'text-to-words')
utt["file_id"] = uttname
utt = complete_utt_from_textgrid(voice, sc_utt, utt)
#add waveform to utt:
utt["waveform"] = Waveform(wavfilename)
#save utt...
ttslab.tofile(utt,
os.path.join(output_dir, ".".join([uttname, UTT_EXT])))
def update_wordview(self):
u = self.corpusview.current_utt
words = u.get_relation("SylStructure").as_list()
word = words[self.corpusview.current_wordindex]
try:
prevword = word.prev_item
prevwordname = prevword["name"]
origstartsample = u["waveform"].samplerate * prevword["start"]
synthstartsample = u["lindists"]["utt"]["waveform"].samplerate * prevword["start"]
prevwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex-1]
except TypeError:
prevwordname = "NONE"
origstartsample = 0
synthstartsample = 0
prevwordpronun = ""
wordname = word["name"]
wordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex]
try:
nextword = word.next_item
nextwordname = nextword["name"]
origendsample = u["waveform"].samplerate * nextword["end"]
synthendsample = u["lindists"]["utt"]["waveform"].samplerate * nextword["end"]
nextwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex+1]
except TypeError:
nextwordname = "NONE"
origendsample = len(u["waveform"].samples)
synthendsample = len(u["waveform"].samples)
nextwordpronun = ""
self.label_word1.set_label(prevwordname)
self.label_word2.set_label(wordname)
self.label_word3.set_label(nextwordname)
self.entry_word1.set_text(prevwordpronun)
self.entry_word2.set_text(wordpronun)
self.entry_word3.set_text(nextwordpronun)
self.origwordcontextwav = Waveform()
self.origwordcontextwav.samplerate = u["waveform"].samplerate
self.origwordcontextwav.samples = u["waveform"].samples[origstartsample:origendsample]
origwordcontext_specfig = Figure(dpi=72)
origwordcontext_specplot = origwordcontext_specfig.add_subplot(111)
origwordcontext_specplot.specgram(self.origwordcontextwav.samples,
Fs=self.origwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.origwordcontextwav.samplerate*len(self.origwordcontextwav.samples)))
origwordcontext_speccanvas = FigureCanvasGTK(origwordcontext_specfig)
framecontents = self.frame_wordspecorig.get_children()
if framecontents:
self.frame_wordspecorig.remove(framecontents[0])
self.frame_wordspecorig.add(origwordcontext_speccanvas)
self.synthwordcontextwav = Waveform()
self.synthwordcontextwav.samplerate = u["lindists"]["utt"]["waveform"].samplerate
self.synthwordcontextwav.samples = u["lindists"]["utt"]["waveform"].samples[synthstartsample:synthendsample]
synthwordcontext_specfig = Figure(dpi=72)
synthwordcontext_specplot = synthwordcontext_specfig.add_subplot(111)
synthwordcontext_specplot.specgram(self.synthwordcontextwav.samples,
Fs=self.synthwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.synthwordcontextwav.samplerate*len(self.synthwordcontextwav.samples)))
synthwordcontext_speccanvas = FigureCanvasGTK(synthwordcontext_specfig)
framecontents = self.frame_wordspecsynth.get_children()
if framecontents:
self.frame_wordspecsynth.remove(framecontents[0])
self.frame_wordspecsynth.add(synthwordcontext_speccanvas)
self.statusbar.push(0, "Item: %s/%s (Word index: %s)" % (self.corpusview.current_index + 1, len(self.corpusview.worklist), self.corpusview.current_wordindex))
self.table_words.show_all()
class SpeechbrowserApp(object):
def __init__(self, voice):
builder = gtk.Builder()
builder.add_from_file(os.path.join(os.path.dirname(__file__), "speechbrowser.glade"))
builder.connect_signals({"on_window1_destroy": gtk.main_quit,
"on_toolbutton_open_clicked": self.on_toolbutton_open_clicked,
"on_button_playutt_clicked": self.on_button_playutt_clicked,
"on_button_playwordorig_clicked": self.on_button_playwordorig_clicked,
"on_button_playwordsynth_clicked": self.on_button_playwordsynth_clicked,
"on_button_next_clicked": self.on_button_next_clicked,
"on_button_prev_clicked": self.on_button_prev_clicked})
self.window1 = builder.get_object("window1")
self.frame_specutt = builder.get_object("frame_specutt")
self.button_playutt = builder.get_object("button_playutt")
self.frame_words = builder.get_object("frame_words")
self.entry_transcription = builder.get_object("entry_transcription")
self.table_utt = builder.get_object("table_utt")
self.table_words = builder.get_object("table_words")
self.frame_wordspecorig = builder.get_object("frame_wordspecorig")
self.frame_wordspecsynth = builder.get_object("frame_wordspecsynth")
self.button_playwordorig = builder.get_object("button_playwordorig")
self.button_playwordsynth = builder.get_object("button_playwordsynth")
self.label_word1 = builder.get_object("label_word1")
self.label_word2 = builder.get_object("label_word2")
self.label_word3 = builder.get_object("label_word3")
self.entry_word1 = builder.get_object("entry_word1")
self.entry_word2 = builder.get_object("entry_word2")
self.entry_word3 = builder.get_object("entry_word3")
self.statusbar = builder.get_object("statusbar")
self.entry_comment = builder.get_object("entry_comment")
# self.combobox_comment = builder.get_object("combobox_comment")
# liststore = gtk.ListStore(gobject.TYPE_STRING)
# self.combobox_comment.set_model(liststore)
# self.combobox_comment.set_entry_text_column(0)
# self.combobox_comment.append_text("transcription error")
# self.combobox_comment.append_text("pronunciation error")
# self.combobox_comment.append_text("noise present")
# self.combobox_comment.append_text("no problem")
# cell = gtk.CellRendererText()
# self.combobox_comment.pack_start(cell, True)
# self.combobox_comment.add_attribute(cell, 'text', 1)
self.window1.show()
self.voice = voice
def update_wordview(self):
u = self.corpusview.current_utt
words = u.get_relation("SylStructure").as_list()
word = words[self.corpusview.current_wordindex]
try:
prevword = word.prev_item
prevwordname = prevword["name"]
origstartsample = u["waveform"].samplerate * prevword["start"]
synthstartsample = u["lindists"]["utt"]["waveform"].samplerate * prevword["start"]
prevwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex-1]
except TypeError:
prevwordname = "NONE"
origstartsample = 0
synthstartsample = 0
prevwordpronun = ""
wordname = word["name"]
wordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex]
try:
nextword = word.next_item
nextwordname = nextword["name"]
origendsample = u["waveform"].samplerate * nextword["end"]
synthendsample = u["lindists"]["utt"]["waveform"].samplerate * nextword["end"]
nextwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex+1]
except TypeError:
nextwordname = "NONE"
origendsample = len(u["waveform"].samples)
synthendsample = len(u["waveform"].samples)
nextwordpronun = ""
self.label_word1.set_label(prevwordname)
self.label_word2.set_label(wordname)
self.label_word3.set_label(nextwordname)
self.entry_word1.set_text(prevwordpronun)
self.entry_word2.set_text(wordpronun)
self.entry_word3.set_text(nextwordpronun)
self.origwordcontextwav = Waveform()
self.origwordcontextwav.samplerate = u["waveform"].samplerate
self.origwordcontextwav.samples = u["waveform"].samples[origstartsample:origendsample]
origwordcontext_specfig = Figure(dpi=72)
origwordcontext_specplot = origwordcontext_specfig.add_subplot(111)
origwordcontext_specplot.specgram(self.origwordcontextwav.samples,
Fs=self.origwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.origwordcontextwav.samplerate*len(self.origwordcontextwav.samples)))
origwordcontext_speccanvas = FigureCanvasGTK(origwordcontext_specfig)
framecontents = self.frame_wordspecorig.get_children()
if framecontents:
self.frame_wordspecorig.remove(framecontents[0])
self.frame_wordspecorig.add(origwordcontext_speccanvas)
self.synthwordcontextwav = Waveform()
self.synthwordcontextwav.samplerate = u["lindists"]["utt"]["waveform"].samplerate
self.synthwordcontextwav.samples = u["lindists"]["utt"]["waveform"].samples[synthstartsample:synthendsample]
synthwordcontext_specfig = Figure(dpi=72)
synthwordcontext_specplot = synthwordcontext_specfig.add_subplot(111)
synthwordcontext_specplot.specgram(self.synthwordcontextwav.samples,
Fs=self.synthwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.synthwordcontextwav.samplerate*len(self.synthwordcontextwav.samples)))
synthwordcontext_speccanvas = FigureCanvasGTK(synthwordcontext_specfig)
framecontents = self.frame_wordspecsynth.get_children()
if framecontents:
self.frame_wordspecsynth.remove(framecontents[0])
self.frame_wordspecsynth.add(synthwordcontext_speccanvas)
self.statusbar.push(0, "Item: %s/%s (Word index: %s)" % (self.corpusview.current_index + 1, len(self.corpusview.worklist), self.corpusview.current_wordindex))
self.table_words.show_all()
def savepronuns(self, wordindex):
if wordindex != 0:
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex-1] = unicode(self.entry_word1.get_text(), "utf-8")
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex] = unicode(self.entry_word2.get_text(), "utf-8")
try:
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex+1] = unicode(self.entry_word3.get_text(), "utf-8")
except IndexError:
pass
def change_wordview(self, button):
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.current_wordindex = button.wordindex
self.update_wordview()
def update_uttview(self):
utt = self.corpusview.current_utt
origspeech_specfig = Figure(dpi=72)
origspeech_specplot = origspeech_specfig.add_subplot(111)
origspeech_specplot.specgram(utt["waveform"].samples, Fs=utt["waveform"].samplerate, NFFT=128, noverlap=64)
origspeech_speccanvas = FigureCanvasGTK(origspeech_specfig)
framecontents = self.frame_specutt.get_children()
if framecontents:
self.frame_specutt.remove(framecontents[0])
self.frame_specutt.add(origspeech_speccanvas)
self.entry_transcription.set_text(self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]])
self.entry_comment.set_text(self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]])
self.buttonbox_words = gtk.HButtonBox()
words = utt.get_relation("Word").as_list()
for i, word in enumerate(words):
button = gtk.Button()
button.wordindex = i
button.connect("clicked", self.change_wordview)
button.set_label(word["name"])
self.buttonbox_words.pack_end(button)
framecontents = self.frame_words.get_children()
if framecontents:
self.frame_words.remove(framecontents[0])
self.frame_words.add(self.buttonbox_words)
self.table_utt.show_all()
self.update_wordview()
def on_button_next_clicked(self, obj):
self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_transcription.get_text(), "utf-8")
self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_comment.get_text(), "utf-8")
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.next()
self.update_uttview()
def on_button_prev_clicked(self, obj):
self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_transcription.get_text(), "utf-8")
self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_comment.get_text(), "utf-8")
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.prev()
self.update_uttview()
def on_button_playutt_clicked(self, obj):
self.corpusview.current_utt["waveform"].play()
def on_button_playwordorig_clicked(self, obj):
self.origwordcontextwav.play()
def on_button_playwordsynth_clicked(self, obj):
self.synthwordcontextwav.play()
def on_toolbutton_open_clicked(self, obj):
chooser = gtk.FileChooserDialog(title=None,
action=gtk.FILE_CHOOSER_ACTION_OPEN,
buttons=(gtk.STOCK_CANCEL,
gtk.RESPONSE_CANCEL,
gtk.STOCK_OPEN,
gtk.RESPONSE_OK))
chooser.set_current_folder(os.getcwd())
response = chooser.run()
if response == gtk.RESPONSE_OK:
filename = chooser.get_filename()
worklist = loadworklist(filename)
self.corpusview = CorpusView(worklist, self.voice)
elif response == gtk.RESPONSE_CANCEL:
print('Closed, no files selected')
chooser.destroy()
self.update_uttview()
self.update_wordview()
def test2(testfn="samples/data_001d.wav"):
aud = Waveform(testfn)
print(inhaletimeinfo(aud.samples))
class SpeechbrowserApp(object):
def __init__(self, phmap):
builder = gtk.Builder()
builder.add_from_file(os.path.join(os.getenv("TTSLABDEV_ROOT"), "voicetools/speechbrowser", "speechbrowser.glade"))
builder.connect_signals({"on_window1_destroy": gtk.main_quit,
"on_toolbutton_open_clicked": self.on_toolbutton_open_clicked,
"on_button_playutt_clicked": self.on_button_playutt_clicked,
"on_button_playwordorig_clicked": self.on_button_playwordorig_clicked,
"on_button_playwordsynth_clicked": self.on_button_playwordsynth_clicked,
"on_button_next_clicked": self.on_button_next_clicked,
"on_button_prev_clicked": self.on_button_prev_clicked})
self.window1 = builder.get_object("window1")
self.frame_specutt = builder.get_object("frame_specutt")
self.button_playutt = builder.get_object("button_playutt")
self.frame_words = builder.get_object("frame_words")
self.entry_transcription = builder.get_object("entry_transcription")
self.table_utt = builder.get_object("table_utt")
self.table_words = builder.get_object("table_words")
self.frame_wordspecorig = builder.get_object("frame_wordspecorig")
self.frame_wordspecsynth = builder.get_object("frame_wordspecsynth")
self.button_playwordorig = builder.get_object("button_playwordorig")
self.button_playwordsynth = builder.get_object("button_playwordsynth")
self.label_word1 = builder.get_object("label_word1")
self.label_word2 = builder.get_object("label_word2")
self.label_word3 = builder.get_object("label_word3")
self.entry_word1 = builder.get_object("entry_word1")
self.entry_word2 = builder.get_object("entry_word2")
self.entry_word3 = builder.get_object("entry_word3")
self.statusbar = builder.get_object("statusbar")
self.entry_comment = builder.get_object("entry_comment")
# self.combobox_comment = builder.get_object("combobox_comment")
# liststore = gtk.ListStore(gobject.TYPE_STRING)
# self.combobox_comment.set_model(liststore)
# self.combobox_comment.set_entry_text_column(0)
# self.combobox_comment.append_text("transcription error")
# self.combobox_comment.append_text("pronunciation error")
# self.combobox_comment.append_text("noise present")
# self.combobox_comment.append_text("no problem")
# cell = gtk.CellRendererText()
# self.combobox_comment.pack_start(cell, True)
# self.combobox_comment.add_attribute(cell, 'text', 1)
self.window1.show()
self.phmap = phmap
def update_wordview(self):
u = self.corpusview.current_utt
words = u.get_relation("SylStructure").as_list()
word = words[self.corpusview.current_wordindex]
try:
prevword = word.prev_item
prevwordname = prevword["name"]
origstartsample = u["waveform"].samplerate * prevword["start"]
synthstartsample = u["lindists"]["utt"]["waveform"].samplerate * prevword["start"]
prevwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex-1]
except TypeError:
prevwordname = "NONE"
origstartsample = 0
synthstartsample = 0
prevwordpronun = ""
wordname = word["name"]
wordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex]
try:
nextword = word.next_item
nextwordname = nextword["name"]
origendsample = u["waveform"].samplerate * nextword["end"]
synthendsample = u["lindists"]["utt"]["waveform"].samplerate * nextword["end"]
nextwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex+1]
except TypeError:
nextwordname = "NONE"
origendsample = len(u["waveform"].samples)
synthendsample = len(u["waveform"].samples)
nextwordpronun = ""
self.label_word1.set_label(prevwordname)
self.label_word2.set_label(wordname)
self.label_word3.set_label(nextwordname)
self.entry_word1.set_text(prevwordpronun)
self.entry_word2.set_text(wordpronun)
self.entry_word3.set_text(nextwordpronun)
self.origwordcontextwav = Waveform()
self.origwordcontextwav.samplerate = u["waveform"].samplerate
self.origwordcontextwav.samples = u["waveform"].samples[origstartsample:origendsample]
origwordcontext_specfig = Figure(dpi=72)
origwordcontext_specplot = origwordcontext_specfig.add_subplot(111)
origwordcontext_specplot.specgram(self.origwordcontextwav.samples,
Fs=self.origwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.origwordcontextwav.samplerate*len(self.origwordcontextwav.samples)))
origwordcontext_speccanvas = FigureCanvasGTK(origwordcontext_specfig)
framecontents = self.frame_wordspecorig.get_children()
if framecontents:
self.frame_wordspecorig.remove(framecontents[0])
self.frame_wordspecorig.add(origwordcontext_speccanvas)
self.synthwordcontextwav = Waveform()
self.synthwordcontextwav.samplerate = u["lindists"]["utt"]["waveform"].samplerate
self.synthwordcontextwav.samples = u["lindists"]["utt"]["waveform"].samples[synthstartsample:synthendsample]
synthwordcontext_specfig = Figure(dpi=72)
synthwordcontext_specplot = synthwordcontext_specfig.add_subplot(111)
synthwordcontext_specplot.specgram(self.synthwordcontextwav.samples,
Fs=self.synthwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.synthwordcontextwav.samplerate*len(self.synthwordcontextwav.samples)))
synthwordcontext_speccanvas = FigureCanvasGTK(synthwordcontext_specfig)
framecontents = self.frame_wordspecsynth.get_children()
if framecontents:
self.frame_wordspecsynth.remove(framecontents[0])
self.frame_wordspecsynth.add(synthwordcontext_speccanvas)
self.statusbar.push(0, "Item: %s/%s (Word index: %s)" % (self.corpusview.current_index + 1, len(self.corpusview.worklist), self.corpusview.current_wordindex))
self.table_words.show_all()
def savepronuns(self, wordindex):
if wordindex != 0:
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex-1] = unicode(self.entry_word1.get_text(), "utf-8")
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex] = unicode(self.entry_word2.get_text(), "utf-8")
try:
self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][wordindex+1] = unicode(self.entry_word3.get_text(), "utf-8")
except IndexError:
pass
def change_wordview(self, button):
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.current_wordindex = button.wordindex
self.update_wordview()
def update_uttview(self):
utt = self.corpusview.current_utt
origspeech_specfig = Figure(dpi=72)
origspeech_specplot = origspeech_specfig.add_subplot(111)
origspeech_specplot.specgram(utt["waveform"].samples, Fs=utt["waveform"].samplerate, NFFT=128, noverlap=64)
origspeech_speccanvas = FigureCanvasGTK(origspeech_specfig)
framecontents = self.frame_specutt.get_children()
if framecontents:
self.frame_specutt.remove(framecontents[0])
self.frame_specutt.add(origspeech_speccanvas)
self.entry_transcription.set_text(self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]])
self.entry_comment.set_text(self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]])
self.buttonbox_words = gtk.HButtonBox()
words = utt.get_relation("Word").as_list()
for i, word in enumerate(words):
button = gtk.Button()
button.wordindex = i
button.connect("clicked", self.change_wordview)
button.set_label(word["name"])
self.buttonbox_words.pack_end(button)
framecontents = self.frame_words.get_children()
if framecontents:
self.frame_words.remove(framecontents[0])
self.frame_words.add(self.buttonbox_words)
self.table_utt.show_all()
self.update_wordview()
def on_button_next_clicked(self, obj):
self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_transcription.get_text(), "utf-8")
self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_comment.get_text(), "utf-8")
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.next()
self.update_uttview()
def on_button_prev_clicked(self, obj):
self.corpusview.transcriptions[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_transcription.get_text(), "utf-8")
self.corpusview.comments[self.corpusview.worklist[self.corpusview.current_index][0]] = unicode(self.entry_comment.get_text(), "utf-8")
self.savepronuns(self.corpusview.current_wordindex)
self.corpusview.prev()
self.update_uttview()
def on_button_playutt_clicked(self, obj):
self.corpusview.current_utt["waveform"].play()
def on_button_playwordorig_clicked(self, obj):
self.origwordcontextwav.play()
def on_button_playwordsynth_clicked(self, obj):
self.synthwordcontextwav.play()
def on_toolbutton_open_clicked(self, obj):
chooser = gtk.FileChooserDialog(title=None,
action=gtk.FILE_CHOOSER_ACTION_OPEN,
buttons=(gtk.STOCK_CANCEL,
gtk.RESPONSE_CANCEL,
gtk.STOCK_OPEN,
gtk.RESPONSE_OK))
chooser.set_current_folder(os.getcwd())
response = chooser.run()
if response == gtk.RESPONSE_OK:
filename = chooser.get_filename()
worklist = loadworklist(filename)
self.corpusview = CorpusView(worklist, self.phmap)
elif response == gtk.RESPONSE_CANCEL:
print('Closed, no files selected')
chooser.destroy()
self.update_uttview()
self.update_wordview()
def update_wordview(self):
u = self.corpusview.current_utt
words = u.get_relation("SylStructure").as_list()
word = words[self.corpusview.current_wordindex]
try:
prevword = word.prev_item
prevwordname = prevword["name"]
origstartsample = u["waveform"].samplerate * prevword["start"]
synthstartsample = u["lindists"]["utt"]["waveform"].samplerate * prevword["start"]
prevwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex-1]
except TypeError:
prevwordname = "NONE"
origstartsample = 0
synthstartsample = 0
prevwordpronun = ""
wordname = word["name"]
wordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex]
try:
nextword = word.next_item
nextwordname = nextword["name"]
origendsample = u["waveform"].samplerate * nextword["end"]
synthendsample = u["lindists"]["utt"]["waveform"].samplerate * nextword["end"]
nextwordpronun = self.corpusview.pronuns[self.corpusview.worklist[self.corpusview.current_index][0]][self.corpusview.current_wordindex+1]
except TypeError:
nextwordname = "NONE"
origendsample = len(u["waveform"].samples)
synthendsample = len(u["waveform"].samples)
nextwordpronun = ""
self.label_word1.set_label(prevwordname)
self.label_word2.set_label(wordname)
self.label_word3.set_label(nextwordname)
self.entry_word1.set_text(prevwordpronun)
self.entry_word2.set_text(wordpronun)
self.entry_word3.set_text(nextwordpronun)
self.origwordcontextwav = Waveform()
self.origwordcontextwav.samplerate = u["waveform"].samplerate
self.origwordcontextwav.samples = u["waveform"].samples[origstartsample:origendsample]
origwordcontext_specfig = Figure(dpi=72)
origwordcontext_specplot = origwordcontext_specfig.add_subplot(111)
origwordcontext_specplot.specgram(self.origwordcontextwav.samples,
Fs=self.origwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.origwordcontextwav.samplerate*len(self.origwordcontextwav.samples)))
origwordcontext_speccanvas = FigureCanvasGTK(origwordcontext_specfig)
framecontents = self.frame_wordspecorig.get_children()
if framecontents:
self.frame_wordspecorig.remove(framecontents[0])
self.frame_wordspecorig.add(origwordcontext_speccanvas)
self.synthwordcontextwav = Waveform()
self.synthwordcontextwav.samplerate = u["lindists"]["utt"]["waveform"].samplerate
self.synthwordcontextwav.samples = u["lindists"]["utt"]["waveform"].samples[synthstartsample:synthendsample]
synthwordcontext_specfig = Figure(dpi=72)
synthwordcontext_specplot = synthwordcontext_specfig.add_subplot(111)
synthwordcontext_specplot.specgram(self.synthwordcontextwav.samples,
Fs=self.synthwordcontextwav.samplerate,
NFFT=128, noverlap=64,
xextent=(0.0, self.synthwordcontextwav.samplerate*len(self.synthwordcontextwav.samples)))
synthwordcontext_speccanvas = FigureCanvasGTK(synthwordcontext_specfig)
framecontents = self.frame_wordspecsynth.get_children()
if framecontents:
self.frame_wordspecsynth.remove(framecontents[0])
self.frame_wordspecsynth.add(synthwordcontext_speccanvas)
self.statusbar.push(0, "Item: %s/%s (Word index: %s)" % (self.corpusview.current_index + 1, len(self.corpusview.worklist), self.corpusview.current_wordindex))
self.table_words.show_all()
def hts_synth(self, utt, processname):
htsparms = self.engine_parms.copy()
htsparms["-of"] = "%(tempolf0_file)s"
if "htsparms" in utt:
htsparms.update(utt["htsparms"]) #parm overrides for this utt...
#build command string and execute:
cmds = self.hts_bin
for k in htsparms:
if htsparms[k]:
if htsparms[k] is True:
cmds += " " + k
else:
cmds += " " + k + " " + str(htsparms[k])
cmds += " %(tempilab_file)s"
fd1, tempwav_file = mkstemp(prefix="ttslab_", suffix=".wav")
fd2, tempilab_file = mkstemp(prefix="ttslab_")
fd3, tempolab_file = mkstemp(prefix="ttslab_")
fd4, tempolf0_file = mkstemp(prefix="ttslab_")
cmds = cmds % {'models_dir': self.models_dir,
'tempwav_file': tempwav_file,
'tempilab_file': tempilab_file,
'tempolab_file': tempolab_file,
'tempolf0_file': tempolf0_file}
#print(cmds)
with codecs.open(tempilab_file, "w", encoding="utf-8") as outfh:
outfh.write("\n".join(utt["hts_label"]))
os.system(cmds)
#load seg endtimes into utt:
with open(tempolab_file) as infh:
lines = infh.readlines()
segs = utt.get_relation("Segment").as_list()
assert len(segs) == len(lines)
for line, seg in zip(lines, segs):
seg["end"] = hts_labels_tone.htk_int_to_float(line.split()[1])
#load audio:
utt["waveform"] = Waveform(tempwav_file)
#load lf0:
f0 = np.exp(np.fromfile(tempolf0_file, "float32")) #load and lf0 to hertz
#to semitones relative to 1Hz:
f0[f0.nonzero()] = 12.0 * np.log2(f0[f0.nonzero()]) # 12 * log2 (F0 / F0reference) where F0reference = 1
f0t = Track()
f0t.values = f0
f0t.times = np.arange(len(f0), dtype=np.float64) * 0.005
utt["f0"] = f0t
#cleanup tempfiles:
os.close(fd1)
os.close(fd2)
os.close(fd3)
os.close(fd4)
os.remove(tempwav_file)
os.remove(tempolab_file)
os.remove(tempilab_file)
os.remove(tempolf0_file)
return utt
