117 lines
3.7 KiB
C++
117 lines
3.7 KiB
C++
// sherpa-onnx/csrc/sherpa-onnx-offline-tts.cc
|
|
//
|
|
// Copyright (c) 2023 Xiaomi Corporation
|
|
|
|
#include <chrono> // NOLINT
|
|
#include <fstream>
|
|
|
|
#include "sherpa-onnx/csrc/offline-tts.h"
|
|
#include "sherpa-onnx/csrc/parse-options.h"
|
|
#include "sherpa-onnx/csrc/wave-writer.h"
|
|
|
|
int32_t audioCallback(const float * /*samples*/, int32_t n, float progress) {
|
|
printf("sample=%d, progress=%f\n", n, progress);
|
|
return 1;
|
|
}
|
|
|
|
int main(int32_t argc, char *argv[]) {
|
|
const char *kUsageMessage = R"usage(
|
|
Offline text-to-speech with sherpa-onnx
|
|
|
|
Usage example:
|
|
|
|
wget https://github.com/k2-fsa/sherpa-onnx/releases/download/tts-models/vits-piper-en_US-amy-low.tar.bz2
|
|
tar xf vits-piper-en_US-amy-low.tar.bz2
|
|
|
|
./bin/sherpa-onnx-offline-tts \
|
|
--vits-model=./vits-piper-en_US-amy-low/en_US-amy-low.onnx \
|
|
--vits-tokens=./vits-piper-en_US-amy-low/tokens.txt \
|
|
--vits-data-dir=./vits-piper-en_US-amy-low/espeak-ng-data \
|
|
--output-filename=./generated.wav \
|
|
"Today as always, men fall into two groups: slaves and free men. Whoever does not have two-thirds of his day for himself, is a slave, whatever he may be: a statesman, a businessman, an official, or a scholar."
|
|
|
|
It will generate a file ./generated.wav as specified by --output-filename.
|
|
|
|
You can find more models at
|
|
https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models
|
|
|
|
Please see
|
|
https://k2-fsa.github.io/sherpa/onnx/tts/index.html
|
|
or details.
|
|
)usage";
|
|
|
|
sherpa_onnx::ParseOptions po(kUsageMessage);
|
|
std::string output_filename = "./generated.wav";
|
|
int32_t sid = 0;
|
|
|
|
po.Register("output-filename", &output_filename,
|
|
"Path to save the generated audio");
|
|
|
|
po.Register("sid", &sid,
|
|
"Speaker ID. Used only for multi-speaker models, e.g., models "
|
|
"trained using the VCTK dataset. Not used for single-speaker "
|
|
"models, e.g., models trained using the LJSpeech dataset");
|
|
|
|
sherpa_onnx::OfflineTtsConfig config;
|
|
|
|
config.Register(&po);
|
|
po.Read(argc, argv);
|
|
|
|
if (po.NumArgs() == 0) {
|
|
fprintf(stderr, "Error: Please provide the text to generate audio.\n\n");
|
|
po.PrintUsage();
|
|
exit(EXIT_FAILURE);
|
|
}
|
|
|
|
if (po.NumArgs() > 1) {
|
|
fprintf(stderr,
|
|
"Error: Accept only one positional argument. Please use single "
|
|
"quotes to wrap your text\n");
|
|
po.PrintUsage();
|
|
exit(EXIT_FAILURE);
|
|
}
|
|
|
|
if (!config.Validate()) {
|
|
fprintf(stderr, "Errors in config!\n");
|
|
exit(EXIT_FAILURE);
|
|
}
|
|
|
|
sherpa_onnx::OfflineTts tts(config);
|
|
|
|
const auto begin = std::chrono::steady_clock::now();
|
|
auto audio = tts.Generate(po.GetArg(1), sid, 1.0, audioCallback);
|
|
const auto end = std::chrono::steady_clock::now();
|
|
|
|
if (audio.samples.empty()) {
|
|
fprintf(
|
|
stderr,
|
|
"Error in generating audio. Please read previous error messages.\n");
|
|
exit(EXIT_FAILURE);
|
|
}
|
|
|
|
float elapsed_seconds =
|
|
std::chrono::duration_cast<std::chrono::milliseconds>(end - begin)
|
|
.count() /
|
|
1000.;
|
|
float duration = audio.samples.size() / static_cast<float>(audio.sample_rate);
|
|
|
|
float rtf = elapsed_seconds / duration;
|
|
fprintf(stderr, "Elapsed seconds: %.3f s\n", elapsed_seconds);
|
|
fprintf(stderr, "Audio duration: %.3f s\n", duration);
|
|
fprintf(stderr, "Real-time factor (RTF): %.3f/%.3f = %.3f\n", elapsed_seconds,
|
|
duration, rtf);
|
|
|
|
bool ok = sherpa_onnx::WriteWave(output_filename, audio.sample_rate,
|
|
audio.samples.data(), audio.samples.size());
|
|
if (!ok) {
|
|
fprintf(stderr, "Failed to write wave to %s\n", output_filename.c_str());
|
|
exit(EXIT_FAILURE);
|
|
}
|
|
|
|
fprintf(stderr, "The text is: %s. Speaker ID: %d\n", po.GetArg(1).c_str(),
|
|
sid);
|
|
fprintf(stderr, "Saved to %s successfully!\n", output_filename.c_str());
|
|
|
|
return 0;
|
|
}
|