1#!/usr/bin/python3 2'''Copyright (c) 2021-2022 Amazon 3 Copyright (c) 2018-2019 Mozilla 4 5 Redistribution and use in source and binary forms, with or without 6 modification, are permitted provided that the following conditions 7 are met: 8 9 - Redistributions of source code must retain the above copyright 10 notice, this list of conditions and the following disclaimer. 11 12 - Redistributions in binary form must reproduce the above copyright 13 notice, this list of conditions and the following disclaimer in the 14 documentation and/or other materials provided with the distribution. 15 16 THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 18 LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 19 A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR 20 CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, 21 EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 22 PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR 23 PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF 24 LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING 25 NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS 26 SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 27''' 28 29# Train an LPCNet model 30 31import argparse 32#from plc_loader import PLCLoader 33 34parser = argparse.ArgumentParser(description='Train a PLC model') 35 36parser.add_argument('features', metavar='<features file>', help='binary features file (float32)') 37parser.add_argument('output', metavar='<output>', help='trained model file (.h5)') 38parser.add_argument('--model', metavar='<model>', default='rdovae', help='PLC model python definition (without .py)') 39group1 = parser.add_mutually_exclusive_group() 40group1.add_argument('--weights', metavar='<input weights>', help='model weights') 41parser.add_argument('--cond-size', metavar='<units>', default=1024, type=int, help='number of units in conditioning network (default 1024)') 42parser.add_argument('--batch-size', metavar='<batch size>', default=1, type=int, help='batch size to use (default 128)') 43parser.add_argument('--seq-length', metavar='<sequence length>', default=1000, type=int, help='sequence length to use (default 1000)') 44 45 46args = parser.parse_args() 47 48import importlib 49rdovae = importlib.import_module(args.model) 50 51from rdovae import apply_dead_zone 52 53import sys 54import numpy as np 55from tensorflow.keras.optimizers import Adam 56from tensorflow.keras.callbacks import ModelCheckpoint, CSVLogger 57import tensorflow.keras.backend as K 58import h5py 59 60import tensorflow as tf 61from rdovae import pvq_quantize 62 63# Try reducing batch_size if you run out of memory on your GPU 64batch_size = args.batch_size 65 66model, encoder, decoder, qembedding = rdovae.new_rdovae_model(nb_used_features=20, nb_bits=80, batch_size=batch_size, cond_size=args.cond_size) 67model.load_weights(args.weights) 68 69lpc_order = 16 70 71feature_file = args.features 72nb_features = model.nb_used_features + lpc_order 73nb_used_features = model.nb_used_features 74sequence_size = args.seq_length 75 76# u for unquantised, load 16 bit PCM samples and convert to mu-law 77 78 79features = np.memmap(feature_file, dtype='float32', mode='r') 80nb_sequences = len(features)//(nb_features*sequence_size)//batch_size*batch_size 81features = features[:nb_sequences*sequence_size*nb_features] 82 83features = np.reshape(features, (nb_sequences, sequence_size, nb_features)) 84print(features.shape) 85features = features[:, :, :nb_used_features] 86#features = np.random.randn(73600, 1000, 17) 87 88 89bits, gru_state_dec = encoder.predict([features], batch_size=batch_size) 90(gru_state_dec).astype('float32').tofile(args.output + "-state.f32") 91 92 93#dist = rdovae.feat_dist_loss(features, quant_out) 94#rate = rdovae.sq1_rate_loss(features, model_bits) 95#rate2 = rdovae.sq_rate_metric(features, model_bits) 96#print(dist, rate, rate2) 97 98print("shapes are:") 99print(bits.shape) 100print(gru_state_dec.shape) 101 102features.astype('float32').tofile(args.output + "-input.f32") 103#quant_out.astype('float32').tofile(args.output + "-enc_dec.f32") 104nbits=80 105bits.astype('float32').tofile(args.output + "-syms.f32") 106 107lambda_val = 0.0002 * np.ones((nb_sequences, sequence_size//2, 1)) 108quant_id = np.round(3.8*np.log(lambda_val/.0002)).astype('int16') 109quant_id = quant_id[:,:,0] 110quant_embed = qembedding(quant_id) 111quant_scale = tf.math.softplus(quant_embed[:,:,:nbits]) 112dead_zone = tf.math.softplus(quant_embed[:, :, nbits : 2 * nbits]) 113 114bits = bits*quant_scale 115bits = np.round(apply_dead_zone([bits, dead_zone]).numpy()) 116bits = bits/quant_scale 117 118gru_state_dec = pvq_quantize(gru_state_dec, 82) 119#gru_state_dec = gru_state_dec/(1e-15+tf.norm(gru_state_dec, axis=-1,keepdims=True)) 120gru_state_dec = gru_state_dec[:,-1,:] 121dec_out = decoder([bits[:,1::2,:], gru_state_dec]) 122 123print(dec_out.shape) 124 125dec_out.numpy().astype('float32').tofile(args.output + "-quant_out.f32") 126