// Run the C++ front end over a text file (one input per line) and write, per line, tab-separated: // phoneme ids (space separated) or "NOTHING" | nasal flags | sentences(finished(text)) joined by U+001F | // shorter(text) joined by U+001F // scripts/tts_hi_g2p_compare.py runs the Python front end on the same lines and reports every difference. // tts_g2p_dump model.gguf texts.txt out.tsv #include #include #include #include "g2p.h" #include "tts.h" static std::string join(const std::vector& v, const std::string& sep) { std::string s; for (size_t i = 0; i < v.size(); ++i) { if (i) s += sep; s += v[i]; } return s; } int main(int argc, char** argv) { if (argc < 4) { std::cerr << "usage: tts_g2p_dump model.gguf texts.txt out.tsv\n"; return 2; } auto model = TtsModel::load(argv[1], 1); HindiG2P g2p(model->symbols()); std::ifstream in(argv[2], std::ios::binary); std::ofstream out(argv[3], std::ios::binary); std::string line; const std::string US = "\x1f"; size_t n = 0; while (std::getline(in, line)) { if (!line.empty() && line.back() == '\r') line.pop_back(); std::string ids, stress; try { const Phonemized p = g2p.phonemize(line); for (size_t i = 0; i < p.ids.size(); ++i) { if (i) { ids += ' '; stress += ' '; } ids += std::to_string(p.ids[i]); stress += std::to_string(p.stress[i]); } } catch (const NothingToPronounce&) { ids = "NOTHING"; } out << ids << '\t' << stress << '\t' << join(split_sentences(finished(line)), US) << '\t' << join(shorter(line), US) << '\n'; ++n; } std::cerr << n << " lines\n"; return 0; }