File size: 8,187 Bytes
4d0da9f | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 | """
Tera.VO - Audio Processing Module
Handles mel spectrogram computation, audio normalization, and vocoder synthesis.
"""
import numpy as np
import tensorflow as tf
from scipy.io import wavfile
from scipy.signal import lfilter
import librosa
import soundfile as sf
import io
class AudioProcessor:
"""Handles all audio processing for Tera.VO TTS system."""
def __init__(
self,
sample_rate=22050,
n_fft=1024,
hop_length=256,
win_length=1024,
n_mels=80,
mel_fmin=0,
mel_fmax=8000,
preemphasis=0.97,
ref_level_db=20,
min_level_db=-100,
max_abs_value=4.0,
griffin_lim_iters=60,
):
self.sample_rate = sample_rate
self.n_fft = n_fft
self.hop_length = hop_length
self.win_length = win_length
self.n_mels = n_mels
self.mel_fmin = mel_fmin
self.mel_fmax = mel_fmax
self.preemphasis = preemphasis
self.ref_level_db = ref_level_db
self.min_level_db = min_level_db
self.max_abs_value = max_abs_value
self.griffin_lim_iters = griffin_lim_iters
# Build mel filter bank
self.mel_basis = librosa.filters.mel(
sr=self.sample_rate,
n_fft=self.n_fft,
n_mels=self.n_mels,
fmin=self.mel_fmin,
fmax=self.mel_fmax,
)
def load_wav(self, path):
"""Load and normalize a wav file."""
wav, sr = librosa.load(path, sr=self.sample_rate)
wav = wav / np.max(np.abs(wav)) * 0.95
return wav
def save_wav(self, wav, path):
"""Save waveform to file."""
wav = wav / np.max(np.abs(wav)) * 0.95
sf.write(path, wav, self.sample_rate)
def preemphasize(self, wav):
"""Apply pre-emphasis filter."""
return lfilter([1, -self.preemphasis], [1], wav)
def inv_preemphasize(self, wav):
"""Inverse pre-emphasis filter."""
return lfilter([1], [1, -self.preemphasis], wav)
def melspectrogram(self, wav):
"""Compute mel spectrogram."""
wav = self.preemphasize(wav)
stft = librosa.stft(
y=wav,
n_fft=self.n_fft,
hop_length=self.hop_length,
win_length=self.win_length,
)
magnitudes = np.abs(stft)
mel = np.dot(self.mel_basis, magnitudes)
mel = self._amp_to_db(mel)
mel = self._normalize(mel)
return mel.T # (T, n_mels)
def inv_melspectrogram(self, mel):
"""Convert mel spectrogram back to waveform using Griffin-Lim."""
mel = mel.T # (n_mels, T)
mel = self._denormalize(mel)
mel = self._db_to_amp(mel)
# Inverse mel filter
inv_mel_basis = np.linalg.pinv(self.mel_basis)
magnitudes = np.maximum(1e-10, np.dot(inv_mel_basis, mel))
# Griffin-Lim reconstruction
wav = self._griffin_lim(magnitudes)
wav = self.inv_preemphasize(wav)
wav = wav / np.max(np.abs(wav)) * 0.95
return wav
def _griffin_lim(self, magnitudes):
"""Griffin-Lim algorithm for phase reconstruction."""
angles = np.exp(2j * np.pi * np.random.rand(*magnitudes.shape))
complex_spec = magnitudes * angles
for i in range(self.griffin_lim_iters):
wav = librosa.istft(
complex_spec,
hop_length=self.hop_length,
win_length=self.win_length,
)
if i < self.griffin_lim_iters - 1:
stft = librosa.stft(
wav,
n_fft=self.n_fft,
hop_length=self.hop_length,
win_length=self.win_length,
)
angles = np.exp(1j * np.angle(stft))
complex_spec = magnitudes * angles
return wav
def _amp_to_db(self, x):
"""Convert amplitude to decibels."""
return 20 * np.log10(np.maximum(1e-5, x)) - self.ref_level_db
def _db_to_amp(self, x):
"""Convert decibels to amplitude."""
return np.power(10.0, (x + self.ref_level_db) * 0.05)
def _normalize(self, S):
"""Normalize spectrogram values."""
return np.clip(
(2 * self.max_abs_value) * ((S - self.min_level_db) / (-self.min_level_db))
- self.max_abs_value,
-self.max_abs_value,
self.max_abs_value,
)
def _denormalize(self, D):
"""Denormalize spectrogram values."""
return (
((np.clip(D, -self.max_abs_value, self.max_abs_value) + self.max_abs_value)
* (-self.min_level_db) / (2 * self.max_abs_value))
+ self.min_level_db
)
def wav_to_bytes(self, wav):
"""Convert waveform array to bytes for Gradio output."""
buffer = io.BytesIO()
sf.write(buffer, wav, self.sample_rate, format='WAV')
buffer.seek(0)
return buffer.read()
class TextProcessor:
"""Handles text normalization and phoneme conversion."""
def __init__(self, max_text_length=200):
self.max_text_length = max_text_length
self._build_char_map()
def _build_char_map(self):
"""Build character to index mapping."""
# Extended character set for better speech synthesis
self.pad = '_'
self.eos = '~'
self.characters = (
self.pad + self.eos
+ 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz'
+ "!'(),.:;? "
+ '-'
+ '0123456789'
)
self.char_to_id = {c: i for i, c in enumerate(self.characters)}
self.id_to_char = {i: c for i, c in enumerate(self.characters)}
self.vocab_size = len(self.characters)
def text_to_sequence(self, text):
"""Convert text string to sequence of character IDs."""
text = self._clean_text(text)
sequence = []
for char in text:
if char in self.char_to_id:
sequence.append(self.char_to_id[char])
sequence.append(self.char_to_id[self.eos])
return sequence
def sequence_to_text(self, sequence):
"""Convert sequence of character IDs back to text."""
return ''.join(self.id_to_char.get(id, '') for id in sequence)
def _clean_text(self, text):
"""Clean and normalize input text."""
try:
from unidecode import unidecode
text = unidecode(text)
except ImportError:
pass
# Basic number to word conversion
text = self._expand_numbers(text)
# Basic abbreviation expansion
text = self._expand_abbreviations(text)
return text
def _expand_numbers(self, text):
"""Expand numbers to words."""
try:
import inflect
p = inflect.engine()
words = text.split()
result = []
for word in words:
stripped = word.strip('.,!?;:')
if stripped.isdigit():
expanded = p.number_to_words(int(stripped))
word = word.replace(stripped, expanded)
result.append(word)
return ' '.join(result)
except (ImportError, Exception):
return text
def _expand_abbreviations(self, text):
"""Expand common abbreviations."""
abbreviations = {
'Mr.': 'Mister',
'Mrs.': 'Misses',
'Dr.': 'Doctor',
'Prof.': 'Professor',
'Jr.': 'Junior',
'Sr.': 'Senior',
'St.': 'Saint',
'etc.': 'et cetera',
'vs.': 'versus',
'i.e.': 'that is',
'e.g.': 'for example',
}
for abbr, full in abbreviations.items():
text = text.replace(abbr, full)
return text
def pad_sequence(self, sequence, max_len=None):
"""Pad sequence to max length."""
if max_len is None:
max_len = self.max_text_length
if len(sequence) >= max_len:
return sequence[:max_len]
return sequence + [0] * (max_len - len(sequence)) |