diff --git a/dnn/dump_data.c b/dnn/dump_data.c
index 053ca774e4332dec17c6ddd0fe9c7f58b779c06b..2c1b783a5b71d8286f15e048235ecf56df2d5b12 100644
--- a/dnn/dump_data.c
+++ b/dnn/dump_data.c
@@ -73,7 +73,7 @@ void compute_noise(int *noise, float noise_std) {
}
}
-static short float2short(float x)
+static opus_int16 float2short(float x)
{
int i;
i = (int)floor(.5+x);
@@ -81,9 +81,9 @@ static short float2short(float x)
}
-void write_audio(LPCNetEncState *st, const short *pcm, const int *noise, FILE *file) {
+void write_audio(LPCNetEncState *st, const opus_int16 *pcm, const int *noise, FILE *file) {
int i;
- short data[2*FRAME_SIZE];
+ opus_int16 data[2*FRAME_SIZE];
for (i=0;i<FRAME_SIZE;i++) {
float p=0;
float e;
@@ -121,9 +121,9 @@ int main(int argc, char **argv) {
FILE *f1;
FILE *ffeat;
FILE *fpcm=NULL;
- short pcm[FRAME_SIZE]={0};
+ opus_int16 pcm[FRAME_SIZE]={0};
int noisebuf[FRAME_SIZE]={0};
- short tmp[FRAME_SIZE] = {0};
+ opus_int16 tmp[FRAME_SIZE] = {0};
float savedX[FRAME_SIZE] = {0};
float speech_gain=1;
int last_silent = 1;
@@ -173,11 +173,11 @@ int main(int argc, char **argv) {
int silent;
size_t ret;
for (i=0;i<FRAME_SIZE;i++) x[i] = tmp[i];
- ret = fread(tmp, sizeof(short), FRAME_SIZE, f1);
+ ret = fread(tmp, sizeof(opus_int16), FRAME_SIZE, f1);
if (feof(f1) || ret != FRAME_SIZE) {
if (!training) break;
rewind(f1);
- ret = fread(tmp, sizeof(short), FRAME_SIZE, f1);
+ ret = fread(tmp, sizeof(opus_int16), FRAME_SIZE, f1);
if (ret != FRAME_SIZE) {
fprintf(stderr, "error reading\n");
exit(1);
@@ -240,7 +240,7 @@ int main(int argc, char **argv) {
process_single_frame(st, ffeat);
if (fpcm) write_audio(st, pcm, noisebuf, fpcm);
- /*if (fpcm) fwrite(pcm, sizeof(short), FRAME_SIZE, fpcm);*/
+ /*if (fpcm) fwrite(pcm, sizeof(opus_int16), FRAME_SIZE, fpcm);*/
for (i=0;i<TRAINING_OFFSET;i++) pcm[i] = float2short(x[i+FRAME_SIZE-TRAINING_OFFSET]);
old_speech_gain = speech_gain;
count++;
diff --git a/dnn/lpcnet.c b/dnn/lpcnet.c
index 1a1b16b798f955e25199553ef29e54df9dfed55b..99a58644630b5c7093a22aab0eddd75ace37a421 100644
--- a/dnn/lpcnet.c
+++ b/dnn/lpcnet.c
@@ -232,7 +232,7 @@ void lpcnet_reset_signal(LPCNetState *lpcnet)
RNN_CLEAR(lpcnet->nnet.gru_b_state, GRU_B_STATE_SIZE);
}
-void lpcnet_synthesize_tail_impl(LPCNetState *lpcnet, short *output, int N, int preload)
+void lpcnet_synthesize_tail_impl(LPCNetState *lpcnet, opus_int16 *output, int N, int preload)
{
int i;
@@ -270,12 +270,12 @@ void lpcnet_synthesize_tail_impl(LPCNetState *lpcnet, short *output, int N, int
}
}
-void lpcnet_synthesize_impl(LPCNetState *lpcnet, const float *features, short *output, int N, int preload)
+void lpcnet_synthesize_impl(LPCNetState *lpcnet, const float *features, opus_int16 *output, int N, int preload)
{
run_frame_network(lpcnet, lpcnet->gru_a_condition, lpcnet->gru_b_condition, lpcnet->lpc, features);
lpcnet_synthesize_tail_impl(lpcnet, output, N, preload);
}
-void lpcnet_synthesize(LPCNetState *lpcnet, const float *features, short *output, int N) {
+void lpcnet_synthesize(LPCNetState *lpcnet, const float *features, opus_int16 *output, int N) {
lpcnet_synthesize_impl(lpcnet, features, output, N, 0);
}
diff --git a/dnn/lpcnet.h b/dnn/lpcnet.h
index 52813c21e4ff7dfadd7cedf2001459a4cf3cb13e..836b168b9bad94130b80ffc9e31ec5c1fa867d98 100644
--- a/dnn/lpcnet.h
+++ b/dnn/lpcnet.h
@@ -27,7 +27,7 @@
#ifndef _LPCNET_H_
#define _LPCNET_H_
-
+#include "opus_types.h"
#define NB_FEATURES 20
#define NB_TOTAL_FEATURES 36
@@ -73,10 +73,10 @@ void lpcnet_decoder_destroy(LPCNetDecState *st);
/** Decodes a packet of LPCNET_COMPRESSED_SIZE bytes (currently 8) into LPCNET_PACKET_SAMPLES samples (currently 640).
* @param [in] st <tt>LPCNetDecState*</tt>: Decoder state
* @param [in] buf <tt>const unsigned char *</tt>: Compressed packet
- * @param [out] pcm <tt>short *</tt>: Decoded audio
+ * @param [out] pcm <tt>opus_int16 *</tt>: Decoded audio
* @retval 0 Success
*/
-int lpcnet_decode(LPCNetDecState *st, const unsigned char *buf, short *pcm);
+int lpcnet_decode(LPCNetDecState *st, const unsigned char *buf, opus_int16 *pcm);
@@ -106,27 +106,19 @@ void lpcnet_encoder_destroy(LPCNetEncState *st);
/** Encodes LPCNET_PACKET_SAMPLES speech samples (currently 640) into a packet of LPCNET_COMPRESSED_SIZE bytes (currently 8).
* @param [in] st <tt>LPCNetDecState*</tt>: Encoder state
- * @param [in] pcm <tt>short *</tt>: Input speech to be encoded
+ * @param [in] pcm <tt>opus_int16 *</tt>: Input speech to be encoded
* @param [out] buf <tt>const unsigned char *</tt>: Compressed packet
* @retval 0 Success
*/
-int lpcnet_encode(LPCNetEncState *st, const short *pcm, unsigned char *buf);
-
-/** Compute features on LPCNET_PACKET_SAMPLES speech samples (currently 640) and output features for 4 10-ms frames at once.
- * @param [in] st <tt>LPCNetDecState*</tt>: Encoder state
- * @param [in] pcm <tt>short *</tt>: Input speech to be analyzed
- * @param [out] features <tt>float[4][NB_TOTAL_FEATURES]</tt>: Four feature vectors
- * @retval 0 Success
- */
-int lpcnet_compute_features(LPCNetEncState *st, const short *pcm, float features[4][NB_TOTAL_FEATURES]);
+int lpcnet_encode(LPCNetEncState *st, const opus_int16 *pcm, unsigned char *buf);
/** Compute features on LPCNET_FRAME_SIZE speech samples (currently 160) and output features for one 10-ms frame.
* @param [in] st <tt>LPCNetDecState*</tt>: Encoder state
- * @param [in] pcm <tt>short *</tt>: Input speech to be analyzed
+ * @param [in] pcm <tt>opus_int16 *</tt>: Input speech to be analyzed
* @param [out] features <tt>float[NB_TOTAL_FEATURES]</tt>: Four feature vectors
* @retval 0 Success
*/
-int lpcnet_compute_single_frame_features(LPCNetEncState *st, const short *pcm, float features[NB_TOTAL_FEATURES]);
+int lpcnet_compute_single_frame_features(LPCNetEncState *st, const opus_int16 *pcm, float features[NB_TOTAL_FEATURES]);
/** Compute features on LPCNET_FRAME_SIZE speech samples (currently 160) and output features for one 10-ms frame.
@@ -164,11 +156,11 @@ void lpcnet_destroy(LPCNetState *st);
/** Synthesizes speech from an LPCNet feature vector.
* @param [in] st <tt>LPCNetState*</tt>: Synthesis state
* @param [in] features <tt>const float *</tt>: Compressed packet
- * @param [out] output <tt>short **</tt>: Synthesized speech
+ * @param [out] output <tt>opus_int16 **</tt>: Synthesized speech
* @param [in] N <tt>int</tt>: Number of samples to generate
* @retval 0 Success
*/
-void lpcnet_synthesize(LPCNetState *st, const float *features, short *output, int N);
+void lpcnet_synthesize(LPCNetState *st, const float *features, opus_int16 *output, int N);
#define LPCNET_PLC_CAUSAL 0
@@ -184,9 +176,9 @@ LPCNetPLCState *lpcnet_plc_create(int options);
void lpcnet_plc_destroy(LPCNetPLCState *st);
-int lpcnet_plc_update(LPCNetPLCState *st, short *pcm);
+int lpcnet_plc_update(LPCNetPLCState *st, opus_int16 *pcm);
-int lpcnet_plc_conceal(LPCNetPLCState *st, short *pcm);
+int lpcnet_plc_conceal(LPCNetPLCState *st, opus_int16 *pcm);
void lpcnet_plc_fec_add(LPCNetPLCState *st, const float *features);
diff --git a/dnn/lpcnet_demo.c b/dnn/lpcnet_demo.c
index 1b1a56505fd4116499a8c3ef2071b7e472909c03..a41f51208e3539c4aab06b00f2d9c940021f3743 100644
--- a/dnn/lpcnet_demo.c
+++ b/dnn/lpcnet_demo.c
@@ -158,7 +158,7 @@ int main(int argc, char **argv) {
net = lpcnet_encoder_create();
while (1) {
float features[NB_TOTAL_FEATURES];
- short pcm[LPCNET_FRAME_SIZE];
+ opus_int16 pcm[LPCNET_FRAME_SIZE];
size_t ret;
ret = fread(pcm, sizeof(pcm[0]), LPCNET_FRAME_SIZE, fin);
if (feof(fin) || ret != LPCNET_FRAME_SIZE) break;
@@ -175,7 +175,7 @@ int main(int argc, char **argv) {
while (1) {
float in_features[NB_TOTAL_FEATURES];
float features[NB_FEATURES];
- short pcm[LPCNET_FRAME_SIZE];
+ opus_int16 pcm[LPCNET_FRAME_SIZE];
size_t ret;
ret = fread(in_features, sizeof(features[0]), NB_TOTAL_FEATURES, fin);
if (feof(fin) || ret != NB_TOTAL_FEATURES) break;
@@ -185,7 +185,7 @@ int main(int argc, char **argv) {
}
lpcnet_destroy(net);
} else if (mode == MODE_PLC) {
- short pcm[FRAME_SIZE];
+ opus_int16 pcm[FRAME_SIZE];
int count=0;
int loss=0;
int skip=0, extra=0;
diff --git a/dnn/lpcnet_enc.c b/dnn/lpcnet_enc.c
index f5e33689f980cfd66a5beb1e039eb68f96c9006e..0e77ced2778df6bbb5d8d97ba7d194f403c667a5 100644
--- a/dnn/lpcnet_enc.c
+++ b/dnn/lpcnet_enc.c
@@ -233,7 +233,7 @@ static int lpcnet_compute_single_frame_features_impl(LPCNetEncState *st, float *
return 0;
}
-int lpcnet_compute_single_frame_features(LPCNetEncState *st, const short *pcm, float features[NB_TOTAL_FEATURES]) {
+int lpcnet_compute_single_frame_features(LPCNetEncState *st, const opus_int16 *pcm, float features[NB_TOTAL_FEATURES]) {
int i;
float x[FRAME_SIZE];
for (i=0;i<FRAME_SIZE;i++) x[i] = pcm[i];
diff --git a/dnn/lpcnet_plc.c b/dnn/lpcnet_plc.c
index 6d384a0370c12a472dd484ee692c2fc8eb188712..6f61b9c74afdaa34abbe02d3b8dd364a9f1fcb5b 100644
--- a/dnn/lpcnet_plc.c
+++ b/dnn/lpcnet_plc.c
@@ -175,10 +175,10 @@ void clear_state(LPCNetPLCState *st) {
/* In this causal version of the code, the DNN model implemented by compute_plc_pred()
needs to generate two feature vectors to conceal the first lost packet.*/
-int lpcnet_plc_update(LPCNetPLCState *st, short *pcm) {
+int lpcnet_plc_update(LPCNetPLCState *st, opus_int16 *pcm) {
int i;
float x[FRAME_SIZE];
- short output[FRAME_SIZE];
+ opus_int16 output[FRAME_SIZE];
float plc_features[2*NB_BANDS+NB_FEATURES+1];
int delta = 0;
for (i=0;i<FRAME_SIZE;i++) x[i] = pcm[i];
@@ -186,7 +186,7 @@ int lpcnet_plc_update(LPCNetPLCState *st, short *pcm) {
if (st->skip_analysis) {
/*fprintf(stderr, "skip update\n");*/
if (st->blend) {
- short tmp[FRAME_SIZE-TRAINING_OFFSET];
+ opus_int16 tmp[FRAME_SIZE-TRAINING_OFFSET];
float zeros[2*NB_BANDS+NB_FEATURES+1] = {0};
RNN_COPY(zeros, plc_features, 2*NB_BANDS);
zeros[2*NB_BANDS+NB_FEATURES] = 1;
@@ -263,9 +263,9 @@ int lpcnet_plc_update(LPCNetPLCState *st, short *pcm) {
}
static const float att_table[10] = {0, 0, -.2, -.2, -.4, -.4, -.8, -.8, -1.6, -1.6};
-int lpcnet_plc_conceal(LPCNetPLCState *st, short *pcm) {
+int lpcnet_plc_conceal(LPCNetPLCState *st, opus_int16 *pcm) {
int i;
- short output[FRAME_SIZE];
+ opus_int16 output[FRAME_SIZE];
run_frame_network_flush(&st->lpcnet);
/* If we concealed the previous frame, finish synthesizing the rest of the samples. */
/* FIXME: Copy/predict features. */
diff --git a/dnn/lpcnet_private.h b/dnn/lpcnet_private.h
index a0f0f7dac1670229440a6f8c57465ef8a190bab0..a2eba6dbd5d6ca0b0e4b5c25402e5c460dd66ede 100644
--- a/dnn/lpcnet_private.h
+++ b/dnn/lpcnet_private.h
@@ -77,7 +77,7 @@ struct LPCNetPLCState {
int fec_read_pos;
int fec_fill_pos;
int fec_skip;
- short pcm[PLC_BUF_SIZE+FRAME_SIZE];
+ opus_int16 pcm[PLC_BUF_SIZE+FRAME_SIZE];
int pcm_fill;
int skip_analysis;
int blend;
@@ -91,19 +91,17 @@ void preemphasis(float *y, float *mem, const float *x, float coef, int N);
void compute_frame_features(LPCNetEncState *st, const float *in);
-void decode_packet(float features[4][NB_TOTAL_FEATURES], float *vq_mem, const unsigned char buf[8]);
-
void lpcnet_reset_signal(LPCNetState *lpcnet);
void run_frame_network(LPCNetState *lpcnet, float *gru_a_condition, float *gru_b_condition, float *lpc, const float *features);
void run_frame_network_deferred(LPCNetState *lpcnet, const float *features);
void run_frame_network_flush(LPCNetState *lpcnet);
-void lpcnet_synthesize_tail_impl(LPCNetState *lpcnet, short *output, int N, int preload);
-void lpcnet_synthesize_impl(LPCNetState *lpcnet, const float *features, short *output, int N, int preload);
-void lpcnet_synthesize_blend_impl(LPCNetState *lpcnet, const short *pcm_in, short *output, int N);
+void lpcnet_synthesize_tail_impl(LPCNetState *lpcnet, opus_int16 *output, int N, int preload);
+void lpcnet_synthesize_impl(LPCNetState *lpcnet, const float *features, opus_int16 *output, int N, int preload);
+void lpcnet_synthesize_blend_impl(LPCNetState *lpcnet, const opus_int16 *pcm_in, opus_int16 *output, int N);
void process_single_frame(LPCNetEncState *st, FILE *ffeat);
-int lpcnet_compute_single_frame_features(LPCNetEncState *st, const short *pcm, float features[NB_TOTAL_FEATURES]);
+int lpcnet_compute_single_frame_features(LPCNetEncState *st, const opus_int16 *pcm, float features[NB_TOTAL_FEATURES]);
void process_single_frame(LPCNetEncState *st, FILE *ffeat);