option to save tts as mp3, currently slightly bugged

embd_res/kcpp_musicui.embd

@@ -307,6 +307,11 @@ select{
     <input title="TTS Instruction" id="tts_instruction" placeholder="e.g. angry shouting loud male">
   </div>
 
+  <div style="margin-top:10px">
+    <label>Save as MP3</label>
+    <input title="Save as MP3" id="tts_use_mp3" type="checkbox" style="max-width:30px">
+  </div>
+
   <div style="margin-top:14px">
     <label>API Base URL (optional)</label>
     <input id="tts_baseUrl" placeholder="http://localhost:5001">
@@ -445,7 +450,8 @@ async function generateTTS(){
 
     const payload = {
       input: document.getElementById("tts_input").value,
-      voice: document.getElementById("tts_voice").value
+      voice: document.getElementById("tts_voice").value,
+      use_mp3: document.getElementById("tts_use_mp3").checked
     };
 
     const instruction = document.getElementById("tts_instruction").value;
@@ -495,6 +501,7 @@ async function generateTTS(){
 function clearTTS(){
   document.getElementById("tts_input").value="";
   document.getElementById("tts_instruction").value="";
+  document.getElementById("tts_use_mp3").checked=false;
 }
 
 //end of tts part
@@ -935,4 +942,4 @@ fetchStats();
 </script>
 
 </body>
-</html>
+</html>

expose.h

@@ -320,6 +320,7 @@ struct tts_generation_inputs
     const char * custom_speaker_data = "";
     const char * reference_audio = "";
     const char * speaker_instruction = "";
+    const bool use_mp3 = false;
 };
 struct tts_generation_outputs
 {

koboldcpp.py

@@ -498,7 +498,8 @@ class tts_generation_inputs(ctypes.Structure):
                 ("custom_speaker_text", ctypes.c_char_p),
                 ("custom_speaker_data", ctypes.c_char_p),
                 ("reference_audio", ctypes.c_char_p),
-                ("speaker_instruction", ctypes.c_char_p)]
+                ("speaker_instruction", ctypes.c_char_p),
+                ("use_mp3", ctypes.c_bool)]
 
 class tts_generation_outputs(ctypes.Structure):
     _fields_ = [("status", ctypes.c_int),
@@ -2999,6 +3000,7 @@ def tts_generate(genparams):
     if not genparams.get("instruction", ""):
         prompt, ttsinstruction = tts_extract_instruction(prompt)
     inputs.speaker_instruction = ttsinstruction.encode("UTF-8")
+    inputs.use_mp3 = genparams.get("use_mp3", False)
     inputs.prompt = prompt.encode("UTF-8")
     inputs.speaker_seed = voice
     aseed = -1

otherarch/tts_adapter.cpp

@@ -492,6 +492,19 @@ static int code_terminate_id = 151670;
 static int nthreads = 4;
 static int tts_max_len = 4096;
 
+static std::string save_tts_audio_base64(const std::vector<float> & audio, int sample_rate, bool use_mp3)
+{
+    if (!use_mp3) {
+        return save_wav16_base64(audio, sample_rate);
+    }
+
+    std::vector<float> stereo_audio;
+    stereo_audio.reserve(audio.size() * 2);
+    stereo_audio.insert(stereo_audio.end(), audio.begin(), audio.end());
+    stereo_audio.insert(stereo_audio.end(), audio.begin(), audio.end());
+    return save_stereo_mp3_base64(stereo_audio, (int) audio.size(), sample_rate);
+}
+
 //ttscpp specific
 static bool is_ttscpp_file = false;
 static generation_configuration * ttscpp_config = nullptr;
@@ -718,7 +731,7 @@ static tts_generation_outputs ttstype_generate_ttscpp(const tts_generation_input
         printf("\nTTS Generated audio in %.2fs.\n",ttstime);
         std::vector<float> wavdat = std::vector(response_data.data, response_data.data + response_data.n_outputs);
         //audio_post_clean(wavdat);
-        last_generated_audio = save_wav16_base64(wavdat, ttscpp_runner->sampling_rate);
+        last_generated_audio = save_tts_audio_base64(wavdat, ttscpp_runner->sampling_rate, inputs.use_mp3);
         output.data = last_generated_audio.c_str();
         output.status = 1;
         last_generation_settings_audio_seed = 0;
@@ -1131,7 +1144,7 @@ static tts_generation_outputs ttstype_generate_outetts(const tts_generation_inpu
             return output;
         }
 
-        last_generated_audio = save_wav16_base64(audio, t_sr);
+        last_generated_audio = save_tts_audio_base64(audio, t_sr, inputs.use_mp3);
         ttstime = timer_check();
 
         printf("\nTTS Generated %d audio tokens in %.2fs.\n",(int) codes.size(),ttstime);
@@ -1239,7 +1252,7 @@ static tts_generation_outputs ttstype_generate_qwen3tts(const tts_generation_inp
 
         ttstime = timer_check();
         printf("\nTTS Generated audio in %.2fs.\n",ttstime);
-        last_generated_audio = save_wav16_base64(result.audio, result.sample_rate);
+        last_generated_audio = save_tts_audio_base64(result.audio, result.sample_rate, inputs.use_mp3);
         output.data = last_generated_audio.c_str();
         output.status = 1;
         last_generation_settings_audio_seed = inputs.audio_seed;