def load_vocoder(self, model_file, model_config, use_cuda):
self.vocoder_config = load_config(model_config)
self.vocoder_model = setup_generator(self.vocoder_config)
self.vocoder_model.load_state_dict(
torch.load(model_file, map_location="cpu")["model"])
self.vocoder_model.remove_weight_norm()
self.vocoder_model.inference_padding = 0
self.vocoder_config = load_config(model_config)
if use_cuda:
self.vocoder_model.cuda()
self.vocoder_model.eval()
def __init__(self, path, INPUT_SR, TARGET_SR, WINDOW_LENGTH):
self.INPUT_SR = INPUT_SR
self.TARGET_SR = TARGET_SR
self.WINDOW_LENGTH = WINDOW_LENGTH
self.CONFIG = load_config('config_fr.json')
self.CONFIG['audio']['sample_rate'] = self.INPUT_SR
self.AP_INPUT = AudioProcessor(**self.CONFIG['audio'])
self.CONFIG['audio']['sample_rate'] = self.TARGET_SR
self.AP_TARGET = AudioProcessor(**self.CONFIG['audio'])
self.files = glob.glob(path + '/**/*.wav', recursive=True)
#If you change your dataset, delete cache.json
if os.path.isfile('./cache.json'):
with open('./cache.json', "r") as json_file:
self.pre_repertoir = json.load(json_file)
else:
print("> Computing wave files length...")
self.pre_repertoir = [
librosa.get_duration(filename=file)
for file in tqdm(self.files)
]
with open('./cache.json', mode="w") as json_file:
json.dump(self.pre_repertoir, json_file)
self.repertoir = [
int(item / WINDOW_LENGTH) for item in self.pre_repertoir
]
self.length = self.get_len()
def load_wavernn(self, lib_path, model_file, model_config, use_cuda):
# TODO: set a function in wavernn code base for model setup and call it here.
sys.path.append(
lib_path) # set this if WaveRNN is not installed globally
#pylint: disable=import-outside-toplevel
from WaveRNN.models.wavernn import Model
print(" > Loading WaveRNN model ...")
print(" | > model config: ", model_config)
print(" | > model file: ", model_file)
self.wavernn_config = load_config(model_config)
# This is the default architecture we use for our models.
# You might need to update it
self.wavernn = Model(
rnn_dims=512,
fc_dims=512,
mode=self.wavernn_config.mode,
mulaw=self.wavernn_config.mulaw,
pad=self.wavernn_config.pad,
use_aux_net=self.wavernn_config.use_aux_net,
use_upsample_net=self.wavernn_config.use_upsample_net,
upsample_factors=self.wavernn_config.upsample_factors,
feat_dims=80,
compute_dims=128,
res_out_dims=128,
res_blocks=10,
hop_length=self.ap.hop_length,
sample_rate=self.ap.sample_rate,
).cuda()
check = torch.load(model_file, map_location="cpu")
self.wavernn.load_state_dict(check['model'])
if use_cuda:
self.wavernn.cuda()
self.wavernn.eval()
def test_in_out(self):
self._create_random_model()
config = load_config(
os.path.join(get_tests_input_path(), 'server_config.json'))
tts_root_path = get_tests_output_path()
config['tts_checkpoint'] = os.path.join(tts_root_path,
config['tts_checkpoint'])
config['tts_config'] = os.path.join(tts_root_path,
config['tts_config'])
synthesizer = Synthesizer(config)
synthesizer.tts("Better this test works!!")
def _create_random_model(self):
# pylint: disable=global-statement
global symbols, phonemes
config = load_config(
os.path.join(get_tests_output_path(), 'dummy_model_config.json'))
if 'characters' in config.keys():
symbols, phonemes = make_symbols(**config.characters)
num_chars = len(phonemes) if config.use_phonemes else len(symbols)
model = setup_model(num_chars, 0, config)
output_path = os.path.join(get_tests_output_path())
save_checkpoint(model, None, 10, 10, 1, output_path)
def load_tts(self, tts_checkpoint, tts_config, use_cuda):
# pylint: disable=global-statement
global symbols, phonemes
print(" > Loading TTS model ...")
print(" | > model config: ", tts_config)
print(" | > checkpoint file: ", tts_checkpoint)
self.tts_config = load_config(tts_config)
self.use_phonemes = self.tts_config.use_phonemes
self.ap = AudioProcessor(**self.tts_config.audio)
if 'characters' in self.tts_config.keys():
symbols, phonemes = make_symbols(**self.tts_config.characters)
if self.use_phonemes:
self.input_size = len(phonemes)
else:
self.input_size = len(symbols)
# TODO: fix this for multi-speaker model - load speakers
if self.config.tts_speakers is not None:
self.tts_speakers = load_speaker_mapping(self.config.tts_speakers)
num_speakers = len(self.tts_speakers)
else:
num_speakers = 0
self.tts_model = setup_model(self.input_size,
num_speakers=num_speakers,
c=self.tts_config)
# load model state
cp = torch.load(tts_checkpoint, map_location=torch.device('cpu'))
# load the model
self.tts_model.load_state_dict(cp['model'])
if use_cuda:
self.tts_model.cuda()
self.tts_model.eval()
self.tts_model.decoder.max_decoder_steps = 3000
if 'r' in cp:
self.tts_model.decoder.set_r(cp['r'])
print(f" > model reduction factor: {cp['r']}")
def main():
"""Run preprocessing process."""
parser = argparse.ArgumentParser(
description="Compute mean and variance of spectrogtram features.")
parser.add_argument(
"--config_path",
type=str,
required=True,
help="TTS config file path to define audio processin parameters.")
parser.add_argument("--out_path",
default=None,
type=str,
help="directory to save the output file.")
args = parser.parse_args()
# load config
CONFIG = load_config(args.config_path)
CONFIG.audio['signal_norm'] = False # do not apply earlier normalization
CONFIG.audio['stats_path'] = None # discard pre-defined stats
# load audio processor
ap = AudioProcessor(**CONFIG.audio)
# load the meta data of target dataset
dataset_items = load_meta_data(CONFIG.datasets)[0] # take only train data
print(f" > There are {len(dataset_items)} files.")
mel_sum = 0
mel_square_sum = 0
linear_sum = 0
linear_square_sum = 0
N = 0
for item in tqdm(dataset_items):
# compute features
wav = ap.load_wav(item[1])
linear = ap.spectrogram(wav)
mel = ap.melspectrogram(wav)
# compute stats
N += mel.shape[1]
mel_sum += mel.sum(1)
linear_sum += linear.sum(1)
mel_square_sum += (mel**2).sum(axis=1)
linear_square_sum += (linear**2).sum(axis=1)
mel_mean = mel_sum / N
mel_scale = np.sqrt(mel_square_sum / N - mel_mean**2)
linear_mean = linear_sum / N
linear_scale = np.sqrt(linear_square_sum / N - linear_mean**2)
output_file_path = os.path.join(args.out_path, "scale_stats.npy")
stats = {}
stats['mel_mean'] = mel_mean
stats['mel_std'] = mel_scale
stats['linear_mean'] = linear_mean
stats['linear_std'] = linear_scale
print(f' > Avg mel spec mean: {mel_mean.mean()}')
print(f' > Avg mel spec scale: {mel_scale.mean()}')
print(f' > Avg linear spec mean: {linear_mean.mean()}')
print(f' > Avg lienar spec scale: {linear_scale.mean()}')
# set default config values for mean-var scaling
CONFIG.audio['stats_path'] = output_file_path
CONFIG.audio['signal_norm'] = True
# remove redundant values
del CONFIG.audio['max_norm']
del CONFIG.audio['min_level_db']
del CONFIG.audio['symmetric_norm']
del CONFIG.audio['clip_norm']
stats['audio_config'] = CONFIG.audio
np.save(output_file_path, stats, allow_pickle=True)
print(f' > scale_stats.npy is saved to {output_file_path}')
default="",
help='DISTRIBUTED: process group id.')
args = parser.parse_args()
if args.continue_path != '':
args.output_path = args.continue_path
args.config_path = os.path.join(args.continue_path, 'config.json')
list_of_files = glob.glob(
args.continue_path +
"/*.pth.tar") # * means all if need specific format then *.csv
latest_model_file = max(list_of_files, key=os.path.getctime)
args.restore_path = latest_model_file
print(f" > Training continues for {args.restore_path}")
# setup output paths and read configs
c = load_config(args.config_path)
# check_config(c)
_ = os.path.dirname(os.path.realpath(__file__))
OUT_PATH = args.continue_path
if args.continue_path == '':
OUT_PATH = create_experiment_folder(c.output_path, c.run_name,
args.debug)
AUDIO_PATH = os.path.join(OUT_PATH, 'test_audios')
c_logger = ConsoleLogger()
if args.rank == 0:
os.makedirs(AUDIO_PATH, exist_ok=True)
new_fields = {}
from tests import get_tests_input_path
from mozilla_voice_tts.tts.tf.models.tacotron2 import Tacotron2
from mozilla_voice_tts.tts.tf.utils.tflite import (convert_tacotron2_to_tflite,
load_tflite_model)
from mozilla_voice_tts.utils.io import load_config
tf.get_logger().setLevel('INFO')
#pylint: disable=unused-variable
torch.manual_seed(1)
use_cuda = torch.cuda.is_available()
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
c = load_config(os.path.join(get_tests_input_path(), 'test_config.json'))
class TacotronTFTrainTest(unittest.TestCase):
@staticmethod
def generate_dummy_inputs():
chars_seq = torch.randint(0, 24, (8, 128)).long().to(device)
chars_seq_lengths = torch.randint(100, 128, (8, )).long().to(device)
chars_seq_lengths = torch.sort(chars_seq_lengths, descending=True)[0]
mel_spec = torch.rand(8, 30, c.audio['num_mels']).to(device)
mel_postnet_spec = torch.rand(8, 30, c.audio['num_mels']).to(device)
mel_lengths = torch.randint(20, 30, (8, )).long().to(device)
stop_targets = torch.zeros(8, 30, 1).float().to(device)
speaker_ids = torch.randint(0, 5, (8, )).long().to(device)
chars_seq = tf.convert_to_tensor(chars_seq.cpu().numpy())
import os
import unittest
import torch as T
from tests import get_tests_input_path
from mozilla_voice_tts.speaker_encoder.losses import GE2ELoss, AngleProtoLoss
from mozilla_voice_tts.speaker_encoder.model import SpeakerEncoder
from mozilla_voice_tts.utils.io import load_config
file_path = get_tests_input_path()
c = load_config(os.path.join(file_path, "test_config.json"))
class SpeakerEncoderTests(unittest.TestCase):
# pylint: disable=R0201
def test_in_out(self):
dummy_input = T.rand(4, 20, 80) # B x T x D
dummy_hidden = [T.rand(2, 4, 128), T.rand(2, 4, 128)]
model = SpeakerEncoder(input_dim=80,
proj_dim=256,
lstm_dim=768,
num_lstm_layers=3)
# computing d vectors
output = model.forward(dummy_input)
assert output.shape[0] == 4
assert output.shape[1] == 256
output = model.inference(dummy_input)
assert output.shape[0] == 4
assert output.shape[1] == 256
# compute d vectors by passing LSTM hidden
parser = argparse.ArgumentParser()
parser.add_argument('--torch_model_path',
type=str,
help='Path to target torch model to be converted to TF.')
parser.add_argument('--config_path',
type=str,
help='Path to config file of torch model.')
parser.add_argument(
'--output_path',
type=str,
help='path to output file including file name to save TF model.')
args = parser.parse_args()
# load model config
config_path = args.config_path
c = load_config(config_path)
num_speakers = 0
# init torch model
model = setup_generator(c)
checkpoint = torch.load(args.torch_model_path,
map_location=torch.device('cpu'))
state_dict = checkpoint['model']
model.load_state_dict(state_dict)
model.remove_weight_norm()
state_dict = model.state_dict()
# init tf model
model_tf = setup_tf_generator(c)
common_sufix = '/.ATTRIBUTES/VARIABLE_VALUE'
help="JSON file for multi-speaker model.",
default="")
parser.add_argument(
'--speaker_fileid',
type=str,
help="if CONFIG.use_external_speaker_embedding_file is true, name of speaker embedding reference file present in speakers.json, else target speaker_fileid if the model is multi-speaker.",
default=None)
parser.add_argument(
'--gst_style',
help="Wav path file for GST stylereference.",
default=None)
args = parser.parse_args()
# load the config
C = load_config(args.config_path)
C.forward_attn_mask = True
# load the audio processor
ap = AudioProcessor(**C.audio)
# if the vocabulary was passed, replace the default
if 'characters' in C.keys():
symbols, phonemes = make_symbols(**C.characters)
speaker_embedding = None
speaker_embedding_dim = None
num_speakers = 0
# load speakers
if args.speakers_json != '':