Cutlery/calibre
1
0
Fork 0
mirror of https://github.com/kovidgoyal/calibre.git synced 2025-08-11 09:13:57 -04:00
Code Issues Packages Projects Releases Wiki Activity

Avoid malloc for normalization

Browse Source
This commit is contained in:
Kovid Goyal 2025-07-30 17:24:59 +05:30
parent c61297560f
commit d835fea08c
No known key found for this signature in database
GPG Key ID: 06BC317B515ACE7C
1 changed files with 9 additions and 14 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

23
src/calibre/utils/tts/piper.cpp
View File

@ -431,32 +431,28 @@ next(PyObject *self, PyObject *args) {
} }
int num_samples; const float *audio_tensor_data; int num_samples; const float *audio_tensor_data;
Py_BEGIN_ALLOW_THREADS; PyObject *ans = NULL, *data = NULL;
int num_of_silence_samples = 0;
auto audio_shape = output_tensors.front().GetTensorTypeAndShapeInfo().GetShape(); auto audio_shape = output_tensors.front().GetTensorTypeAndShapeInfo().GetShape();
num_samples = (int)audio_shape[audio_shape.size() - 1]; num_samples = (int)audio_shape[audio_shape.size() - 1];
audio_tensor_data = output_tensors.front().GetTensorData<float>(); audio_tensor_data = output_tensors.front().GetTensorData<float>();
Py_END_ALLOW_THREADS; float maxval = 1.f;
PyObject *ans = NULL, *data = NULL; Py_BEGIN_ALLOW_THREADS;
int num_of_silence_samples = 0;
if (current_sentence_delay > 0) num_of_silence_samples = (int)(current_sample_rate * current_sentence_delay); if (current_sentence_delay > 0) num_of_silence_samples = (int)(current_sample_rate * current_sentence_delay);
const float *normalized = audio_tensor_data;
if (num_samples) { if (num_samples) {
float maxval = std::abs(audio_tensor_data[0]), q; maxval = std::abs(audio_tensor_data[0]); float q;
for (int i = 1; i < num_samples; i++) if ((q = std::abs(audio_tensor_data[i])) > maxval) maxval = q; for (int i = 1; i < num_samples; i++) if ((q = std::abs(audio_tensor_data[i])) > maxval) maxval = q;
if (maxval > 1e-8) { if (maxval <= 1e-8) maxval = 1.f;
float *temp = (float*)malloc(num_samples * sizeof(audio_tensor_data[0]));
for (int i = 0; i < num_samples; i++) temp[i] /= maxval;
normalized = temp;
}
} }
Py_END_ALLOW_THREADS;
if (as_16bit_samples) { if (as_16bit_samples) {
data = PyBytes_FromStringAndSize(NULL, sizeof(int16_t) * (num_samples + num_of_silence_samples)); data = PyBytes_FromStringAndSize(NULL, sizeof(int16_t) * (num_samples + num_of_silence_samples));
if (data) { if (data) {
Py_BEGIN_ALLOW_THREADS; Py_BEGIN_ALLOW_THREADS;
int16_t *x = (int16_t*)PyBytes_AS_STRING(data); int16_t *x = (int16_t*)PyBytes_AS_STRING(data);
for (int i = 0; i < num_samples; i++) { for (int i = 0; i < num_samples; i++) {
x[i] = (int16_t)(audio_tensor_data[i] * std::numeric_limits<int16_t>::max()); x[i] = (int16_t)((audio_tensor_data[i]/maxval) * std::numeric_limits<int16_t>::max());
} }
memset(x + num_samples, 0, num_of_silence_samples * sizeof(int16_t)); memset(x + num_samples, 0, num_of_silence_samples * sizeof(int16_t));
Py_END_ALLOW_THREADS; Py_END_ALLOW_THREADS;
@ -466,12 +462,11 @@ next(PyObject *self, PyObject *args) {
if (data) { if (data) {
Py_BEGIN_ALLOW_THREADS; Py_BEGIN_ALLOW_THREADS;
float *x = (float*)PyBytes_AS_STRING(data); float *x = (float*)PyBytes_AS_STRING(data);
memcpy(x, audio_tensor_data, sizeof(float) * num_samples); for (int i = 0; i < num_samples; i++) x[i] = audio_tensor_data[i]/maxval;
memset(x + num_samples, 0, num_of_silence_samples * sizeof(int16_t)); memset(x + num_samples, 0, num_of_silence_samples * sizeof(int16_t));
Py_END_ALLOW_THREADS; Py_END_ALLOW_THREADS;
} }
} }
if (normalized != audio_tensor_data) free((void*)normalized);
if (data) { if (data) {
ans = Py_BuildValue( ans = Py_BuildValue(
"OiiO", data, num_samples, current_sample_rate, phoneme_id_queue.empty() ? Py_True : Py_False); "OiiO", data, num_samples, current_sample_rate, phoneme_id_queue.empty() ? Py_True : Py_False);