ref: c45963d40aafc885a68206cda54543688333e1b8
parent: e1181bcad026b0c1f64dacb631a28f833dc21f42
author: Jean-Marc Valin <jmvalin@amazon.com>
date: Mon Jan 31 13:49:19 EST 2022
Code for testing PLC models
--- a/dnn/lpcnet_plc.c
+++ b/dnn/lpcnet_plc.c
@@ -31,6 +31,9 @@
#include "lpcnet_private.h"
#include "lpcnet.h"
+#define PLC_DUMP_FEATURES 0
+#define PLC_READ_FEATURES 0
+
LPCNET_EXPORT int lpcnet_plc_get_size() {return sizeof(LPCNetPLCState);
}
@@ -96,6 +99,13 @@
for (i=0;i<FRAME_SIZE;i++) st->pcm[PLC_BUF_SIZE+i] = pcm[i];
RNN_COPY(output, &st->pcm[0], FRAME_SIZE);
lpcnet_synthesize_impl(&st->lpcnet, st->enc.features[0], output, FRAME_SIZE, FRAME_SIZE);
+#if PLC_READ_FEATURES
+ for (i=0;i<NB_FEATURES;i++) scanf("%f", &st->features[i]);+#endif
+#if PLC_DUMP_FEATURES
+ for (i=0;i<NB_FEATURES;i++) printf("%f ", st->enc.features[0][i]);+ printf("1\n");+#endif
RNN_MOVE(st->pcm, &st->pcm[FRAME_SIZE], PLC_BUF_SIZE);
}
@@ -104,6 +114,9 @@
}
LPCNET_EXPORT int lpcnet_plc_conceal(LPCNetPLCState *st, short *pcm) {+#if PLC_READ_FEATURES || PLC_DUMP_FEATURES
+ int i;
+#endif
short output[FRAME_SIZE];
st->enc.pcount = 0;
/* If we concealed the previous frame, finish synthesizing the rest of the samples. */
@@ -113,7 +126,13 @@
int update_count;
update_count = IMIN(st->pcm_fill, FRAME_SIZE);
RNN_COPY(output, &st->pcm[0], update_count);
-
+#if PLC_READ_FEATURES
+ for (i=0;i<NB_FEATURES;i++) scanf("%f", &st->features[i]);+#endif
+#if PLC_DUMP_FEATURES
+ for (i=0;i<NB_FEATURES+1;i++) printf("%f ", 0.);+ printf("\n");+#endif
lpcnet_synthesize_impl(&st->lpcnet, &st->features[0], output, update_count, update_count);
RNN_MOVE(st->pcm, &st->pcm[FRAME_SIZE], PLC_BUF_SIZE);
st->pcm_fill -= update_count;
@@ -120,6 +139,13 @@
st->skip_analysis++;
}
lpcnet_synthesize_tail_impl(&st->lpcnet, pcm, FRAME_SIZE-TRAINING_OFFSET, 0);
+#if PLC_READ_FEATURES
+ for (i=0;i<NB_FEATURES;i++) scanf("%f", &st->features[i]);+#endif
+#if PLC_DUMP_FEATURES
+ for (i=0;i<NB_FEATURES+1;i++) printf("%f ", 0.);+ printf("\n");+#endif
lpcnet_synthesize_impl(&st->lpcnet, &st->features[0], &pcm[FRAME_SIZE-TRAINING_OFFSET], TRAINING_OFFSET, 0);
{int i;
--- /dev/null
+++ b/dnn/training_tf2/test_plc.py
@@ -1,0 +1,92 @@
+#!/usr/bin/python3
+'''Copyright (c) 2021-2022 Amazon
+ Copyright (c) 2018-2019 Mozilla
+
+ Redistribution and use in source and binary forms, with or without
+ modification, are permitted provided that the following conditions
+ are met:
+
+ - Redistributions of source code must retain the above copyright
+ notice, this list of conditions and the following disclaimer.
+
+ - Redistributions in binary form must reproduce the above copyright
+ notice, this list of conditions and the following disclaimer in the
+ documentation and/or other materials provided with the distribution.
+
+ THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR
+ CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
+ EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
+ PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
+ PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
+ LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
+ NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
+ SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+'''
+
+# Train an LPCNet model
+
+import argparse
+from plc_loader import PLCLoader
+
+parser = argparse.ArgumentParser(description='Test a PLC model')
+
+parser.add_argument('weights', metavar='<weights file>', help='weights file (.h5)')+parser.add_argument('features', metavar='<features file>', help='binary features file (float32)')+parser.add_argument('output', metavar='<output>', help='reconstructed file (float32)')+parser.add_argument('--model', metavar='<model>', default='lpcnet_plc', help='PLC model python definition (without .py)')+group1 = parser.add_mutually_exclusive_group()
+
+parser.add_argument('--gru-size', metavar='<units>', default=256, type=int, help='number of units in GRU (default 256)')+parser.add_argument('--cond-size', metavar='<units>', default=128, type=int, help='number of units in conditioning network (default 128)')+
+
+args = parser.parse_args()
+
+import importlib
+lpcnet = importlib.import_module(args.model)
+
+import sys
+import numpy as np
+from tensorflow.keras.optimizers import Adam
+from tensorflow.keras.callbacks import ModelCheckpoint, CSVLogger
+import tensorflow.keras.backend as K
+import h5py
+
+import tensorflow as tf
+#gpus = tf.config.experimental.list_physical_devices('GPU')+#if gpus:
+# try:
+# tf.config.experimental.set_virtual_device_configuration(gpus[0], [tf.config.experimental.VirtualDeviceConfiguration(memory_limit=5120)])
+# except RuntimeError as e:
+# print(e)
+
+model = lpcnet.new_lpcnet_plc_model(rnn_units=args.gru_size, batch_size=1, training=False, quantize=False, cond_size=args.cond_size)
+model.compile()
+
+lpc_order = 16
+
+feature_file = args.features
+nb_features = model.nb_used_features + lpc_order
+nb_used_features = model.nb_used_features
+
+# u for unquantised, load 16 bit PCM samples and convert to mu-law
+
+features = np.loadtxt(feature_file)
+print(features.shape)
+sequence_size = features.shape[0]
+lost = np.reshape(features[:,-1:], (1, sequence_size, 1))
+features = features[:,:nb_used_features]
+features = np.reshape(features, (1, sequence_size, nb_used_features))
+
+
+model.load_weights(args.weights)
+
+features = features*lost
+out = model.predict([features, lost])
+
+out = features + (1-lost)*out
+
+np.savetxt(args.output, out[0,:,:])
--
⑨