mtmd: support tiny-voca (#32427)

* mtmd: support tiny-voca

* improve G2P
This commit is contained in:
Xuan-Son Nguyen authored and GitHub committed 2026-11-18 00:04:15 +01:00
1 parent 2e53a0bab7
commit 3574df84b7
32 files changed
+4591 -59

No files matched your search

+5 -1
View File
@@ -6986,7 +6986,11 @@ std::unique_ptr<server_res_generator> server_routes::handle_tts_impl(const serve
const std::string voice = json_value(body, "voice", has_speaker_ref ? std::string() : params.tts_voice);
// raw_prompt: the input goes to the model as is (e.g. phonemes), the "tts" chat template is skipped
const bool raw_prompt = json_value(body, "raw_prompt", false);
const std::string tts_prompt = raw_prompt ? std::string() : common_chat_get_tts_prompt(meta->chat_params.tmpls.get(), prompt, lang, voice);
const auto g2p = [this](const std::string & word) {
char buf[512];
return mtmd_gen_audio_g2p(ctx_server.mctx, word.c_str(), buf, sizeof(buf)) > 0 ? std::string(buf) : std::string();
};
const std::string tts_prompt = raw_prompt ? std::string() : common_chat_get_tts_prompt(meta->chat_params.tmpls.get(), prompt, lang, voice, g2p);
// a model with predefined voices only speaks with one of them
mtmd::bitmap_ptr voice_ref;
if (!voice.empty() && info.n_voices > 0) {