def synthesize(self, phonemes_list : List[str]) -> bytes:
"""converts a list of phonemes (only their names) to a wav file"""
pho_list = PhonemeList([])
for pho in phonemes_list:
if pho in self.voice.phonems.VOWELS:
new_phoneme = Phoneme(name=pho,
duration=int(torch.normal(*self.stats.consonants_stats[pho]["duration"])[0]),
pitch_mods=self.gen_vowel_pitches(pho))
else: #it's a consonant
new_phoneme = Phoneme(name=pho,
duration=int(torch.normal(*self.stats.consonants_stats[pho]["duration"])[0]),
pitch_mods=[])
pho_list.append(new_phoneme)
return self.voice.to_audio(pho_list)
def process(self, phonems: PhonemeList):
reconstructed = PhonemeList([])
ng_phonem = Phoneme("N", 100)
euh_phonem = Phoneme("2", 79)
phonems.append(Phoneme("_", 10)) # end-silence-padding, just to be safe
for i, phoneme in enumerate(phonems):
if phoneme.name in FrenchPhonemes.NASAL_WOVELS:
reconstructed += [phoneme, ng_phonem]
elif phoneme.name in FrenchPhonemes.CONSONANTS - {"w"} and phonems[i + 1].name not in FrenchPhonemes.VOWELS:
reconstructed += [phoneme, euh_phonem]
elif phoneme.name == "o":
phoneme.name = "O"
reconstructed.append(phoneme)
else:
reconstructed.append(phoneme)
return reconstructed
def process(self, phonems: PhonemeList):
w_phonem = Phoneme("w", 103, [])
for i, phoneme in enumerate(phonems):
if phoneme.name == "R":
phoneme.name = "w"
if random.randint(0, 1) == 0:
for j in range(2):
phonems.insert(i, w_phonem)
else:
phoneme.duration = 206
return phonems
def process(self, phonems: PhonemeList):
silence = Phoneme("_", 61)
reconstructed = PhonemeList([])
for i, phoneme in enumerate(phonems):
if phonems[i].name in FrenchPhonemes.CONSONANTS \
and phonems[i + 1].name in FrenchPhonemes.VOWELS \
and random.randint(1, 3) == 1:
reconstructed += [phonems[i], phonems[i + 1]] * 2
elif phoneme.name in FrenchPhonemes.VOWELS and random.randint(1, 3) == 1:
reconstructed += [phoneme, silence, phoneme]
else:
reconstructed.append(phoneme)
return reconstructed
beat_duration = len(beats_track) / (rate * BEATS_TRACK_BEAT_COUNT)
beats_track_looped = np.tile(
beats_track,
LOOPS_COUNT * BEATS_PER_MEASURE * len(CHORDS_PROGRESSION) //
BEATS_TRACK_BEAT_COUNT)
logging.info("Beat time : %dms" % (beat_duration * 1000))
logging.info("Measure time : %dms" % (beat_duration * 1000 * 4))
prog_freqs = get_trinote_progression_freqs(CHORDS_PROGRESSION)
logging.info("First freq progression: \n %s" % (str(prog_freqs)))
track_freqs = prog_freqs * LOOPS_COUNT
progression_phonems = PhonemeList([])
for freq in track_freqs:
progression_phonems.append(
Phoneme("a", int(beat_duration * 1000), [(0, freq), (100, freq)]))
logging.info("Rendering audio")
voice = Voice(lang="fr", voice_id=2)
wav = voice.to_audio(progression_phonems)
if BEATS_TRACK_FILE is not None:
rate, wave_array = to_f32_16k(wav)
mixed_tracks = mix_tracks(beats_track_looped * 0.6,
wave_array * 0.3,
align="left")
wav = to_wav_bytes(mixed_tracks, 16000)
player = AudioPlayer()
player.set_file(BytesIO(wav))
player.play()
player.close()