handle contractions correctly, bump defaults

Author: LostRuins

klite.embd

@@ -12,7 +12,7 @@ Current version indicated by LITEVER below.
 -->
 
 <script id="init-config">
-	const LITEVER = 273;
+	const LITEVER = 274;
 	const urlParams = new URLSearchParams(window.location.search);
 	var localflag = urlParams.get('local'); //this will be replaced automatically in embedded kcpp
 	const STORAGE_PREFIX = (localflag?"e_":"")+"kaihordewebui_";
@@ -3348,6 +3348,7 @@ Current version indicated by LITEVER below.
 		notify_on: false,
 		narrate_both_sides: false,
 		narrate_only_dialog: false,
+		embed_narrations: false,
 		voice_end_delay: 300,
 		voice_suppress_nonspeech: false,
 		voice_langcode: "auto",
@@ -3431,7 +3432,7 @@ Current version indicated by LITEVER below.
 		second_ep_url:"",
 
 		max_context_length: (localflag?4096:3072),
-		max_length: (localflag?512:256),
+		max_length: (localflag?640:400),
 		last_maxctx: 0,
 		auto_ctxlen: true,
 		auto_genamt: true,
@@ -12941,6 +12942,7 @@ Current version indicated by LITEVER below.
 		document.getElementById("no_escape_html").checked = no_escape_html;
 		document.getElementById("narrate_both_sides").checked = localsettings.narrate_both_sides;
 		document.getElementById("narrate_only_dialog").checked = localsettings.narrate_only_dialog;
+		document.getElementById("embed_narrations").checked = localsettings.embed_narrations;
 		document.getElementById("tts_speed").value = localsettings.tts_speed;
 		document.getElementById("voice_end_delay").value = localsettings.voice_end_delay;
 		document.getElementById("voice_suppress_nonspeech").checked = localsettings.voice_suppress_nonspeech;
@@ -13473,6 +13475,7 @@ Current version indicated by LITEVER below.
 		no_escape_html = (document.getElementById("no_escape_html").checked?true:false);
 		localsettings.narrate_both_sides = (document.getElementById("narrate_both_sides").checked?true:false);
 		localsettings.narrate_only_dialog = (document.getElementById("narrate_only_dialog").checked?true:false);
+		localsettings.embed_narrations = (document.getElementById("embed_narrations").checked?true:false);
 		localsettings.tts_speed = document.getElementById("tts_speed").value;
 		localsettings.voice_end_delay = document.getElementById("voice_end_delay").value;
 		localsettings.voice_suppress_nonspeech = (document.getElementById("voice_suppress_nonspeech").checked?true:false);
@@ -15448,6 +15451,16 @@ Current version indicated by LITEVER below.
 		a.download = "audio.wav";
 		setTimeout(function(){a.click()},20);
 	}
+	function tts_embed_audio(arrayBufferData)
+	{
+		var file = new Blob([arrayBufferData], { type: 'audio/wav' });
+		var reader = new FileReader();
+		reader.onload = function() {
+			var base64Audio = this.result;
+			self_upload_audio(base64Audio, "tts_audio");
+		};
+		reader.readAsDataURL(file);
+	}
 
 	function tts_speak(text, speech_synth_override=null, do_download=false)
 	{
@@ -15486,6 +15499,12 @@ Current version indicated by LITEVER below.
 			}
 		}
 
+		let do_embed_tts = false;
+		if(!do_download && speech_synth_override==null) //dont use it for tests
+		{
+			do_embed_tts = localsettings.embed_narrations;
+		}
+
 		if(ssval==XTTS_ID || ssval==ALLTALK_ID || ssval==OAI_TTS_ID || ssval==KCPP_TTS_ID || ssval==POLLINATIONS_TTS_ID) //xtts api server
 		{
 			let is_xtts = (ssval==XTTS_ID);
@@ -15544,6 +15563,10 @@ Current version indicated by LITEVER below.
 					{
 						tts_download(audiofile_ref);
 					}
+					if(do_embed_tts)
+					{
+						tts_embed_audio(audiofile_ref);
+					}
 					const playSound = audioContext.createBufferSource();
 					playSound.buffer = decodedData;
 					playSound.connect(audioContext.destination);
@@ -15586,6 +15609,10 @@ Current version indicated by LITEVER below.
 					{
 						tts_download(audiofile_ref);
 					}
+					if(do_embed_tts)
+					{
+						tts_embed_audio(audiofile_ref);
+					}
 					const playSound = audioContext.createBufferSource();
 					playSound.buffer = decodedData;
 					playSound.connect(audioContext.destination);
@@ -15629,6 +15656,10 @@ Current version indicated by LITEVER below.
 						{
 							tts_download(audiofile_ref);
 						}
+						if(do_embed_tts)
+						{
+							tts_embed_audio(audiofile_ref);
+						}
 						const playSound = audioContext.createBufferSource();
 						playSound.buffer = decodedData;
 						playSound.connect(audioContext.destination);
@@ -15692,6 +15723,10 @@ Current version indicated by LITEVER below.
 							{
 								tts_download(audiofile_ref);
 							}
+							if(do_embed_tts)
+							{
+								tts_embed_audio(audiofile_ref);
+							}
 							playDecodedAllTalkData(decodedData);
 						})
 						.catch((error) => {
@@ -15743,6 +15778,10 @@ Current version indicated by LITEVER below.
 											{
 												tts_download(audiofile_ref);
 											}
+											if(do_embed_tts)
+											{
+												tts_embed_audio(audiofile_ref);
+											}
 											playDecodedAllTalkData(decodedData);
 										})
 										.catch((error) => {
@@ -15773,6 +15812,10 @@ Current version indicated by LITEVER below.
 									{
 										tts_download(audiofile_ref);
 									}
+									if(do_embed_tts)
+									{
+										tts_embed_audio(audiofile_ref);
+									}
 									playDecodedAllTalkData(decodedData);
 								}).catch((error) => {
 									console.log("AllTalk v1 Speak Error: " + error);
@@ -20297,13 +20340,20 @@ Current version indicated by LITEVER below.
 
 		if(elements && elements.length>0)
 		{
+			//check if we are on an instruct boundary
+			let onboundary = false;
+			if(gametext_arr.length>0 && gametext_arr[gametext_arr.length-1].trim().endsWith("{{[OUTPUT]}}"))
+			{
+				onboundary = true;
+			}
 			elements.forEach(function (element) {
 				let temp_stream = synchro_pending_stream;
-				// let codeblockcount = (temp_stream.match(/```/g) || []).length;
-				// if(codeblockcount>0 && codeblockcount%2!=0)
-				// {
-				// 	temp_stream += "```"; //force end code block
-				// }
+				if (onboundary) {
+					let codeblockcount = (temp_stream.match(/```/g) || []).length;
+					if (codeblockcount > 0 && codeblockcount % 2 != 0) {
+						temp_stream += "```"; //force end code block
+					}
+				}
 				let pend = escape_html(pending_context_preinjection) + format_streaming_text(escape_html(temp_stream));
 				element.innerHTML = pend;
 			});
@@ -25331,6 +25381,10 @@ Current version indicated by LITEVER below.
 							<div class="justifyleft settingsmall"  title="If unchecked, only speak AI replies, not other text.">Narrate Only Dialog </div>
 							<input title="Narrate Only Dialog" type="checkbox" id="narrate_only_dialog" style="margin:0px 0px 0px auto;">
 						</div>
+						<div class="settinglabel">
+							<div class="justifyleft settingsmall"  title="Try to save and embed TTS narration files into story if possible. Does not work for internal browser TTS. Not recommend due to large size.">Save Narrations </div>
+							<input title="Save Narrations" type="checkbox" id="embed_narrations" style="margin:0px 0px 0px auto;">
+						</div>
 						<div class="inlinelabel" style="font-size: 11px;">
 							<div class="justifyleft">Browser TTS Speed: </div>
 							<input title="Browser Narration Speed" type="text" inputmode="decimal" value="1" id="tts_speed" style="width:40px">

koboldcpp.py

@@ -4524,7 +4524,7 @@ def hide_tooltip(event):
     chatcompletionsadapter_var = ctk.StringVar(value="AutoGuess")
     moeexperts_var = ctk.StringVar(value=str(-1))
     moecpu_var = ctk.StringVar(value=str(0))
-    defaultgenamt_var = ctk.StringVar(value=str(512))
+    defaultgenamt_var = ctk.StringVar(value=str(640))
     nobostoken_var = ctk.IntVar(value=0)
     override_kv_var = ctk.StringVar(value="")
     override_tensors_var = ctk.StringVar(value="")
@@ -5512,7 +5512,7 @@ def export_vars():
             args.overridenativecontext = 0
         args.moeexperts = int(moeexperts_var.get()) if moeexperts_var.get()!="" else -1
         args.moecpu = int(moecpu_var.get()) if moecpu_var.get()!="" else 0
-        args.defaultgenamt = int(defaultgenamt_var.get()) if defaultgenamt_var.get()!="" else 512
+        args.defaultgenamt = int(defaultgenamt_var.get()) if defaultgenamt_var.get()!="" else 640
         args.nobostoken = (nobostoken_var.get()==1)
         args.enableguidance = (enableguidance_var.get()==1)
         args.overridekv = None if override_kv_var.get() == "" else override_kv_var.get()
@@ -7573,7 +7573,7 @@ def range_checker(arg: str):
     advparser.add_argument("--nomodel", help="Allows you to launch the GUI alone, without selecting any model.", action='store_true')
     advparser.add_argument("--moeexperts", metavar=('[num of experts]'), help="How many experts to use for MoE models (default=follow gguf)", type=int, default=-1)
     advparser.add_argument("--moecpu", metavar=('[layers affected]'), help="Keep the Mixture of Experts (MoE) weights of the first N layers in the CPU. If no value is provided, applies to all layers.", nargs='?', const=999, type=int, default=0)
-    advparser.add_argument("--defaultgenamt", help="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.", type=check_range(int,64,8192), default=512)
+    advparser.add_argument("--defaultgenamt", help="How many tokens to generate by default, if not specified. Must be smaller than context size. Usually, your frontend GUI will override this.", type=check_range(int,64,8192), default=640)
     advparser.add_argument("--nobostoken", help="Prevents BOS token from being added at the start of any prompt. Usually NOT recommended for most models.", action='store_true')
     advparser.add_argument("--enableguidance", help="Enables the use of Classifier-Free-Guidance, which allows the use of negative prompts. Has performance and memory impact.", action='store_true')
     advparser.add_argument("--maxrequestsize", metavar=('[size in MB]'), help="Specify a max request payload size. Any requests to the server larger than this size will be dropped. Do not change if unsure.", type=int, default=32)

otherarch/ttscpp/src/kokoro_model.cpp

@@ -1425,8 +1425,9 @@ int kokoro_runner::generate(std::string prompt, struct tts_response * response,
     prompt = replace_any(prompt, ",;:", "--");
     prompt = replace_any(prompt, "\n", " ");
 	kokoro_str_replace_all(prompt," - "," -- ");
+	kokoro_str_replace_all(prompt,"he's ","he is ");
 	kokoro_str_replace_all(prompt,"'s ","s ");
-	kokoro_str_replace_all(prompt,"wasn't ","wasnt ");
+	kokoro_str_replace_all(prompt,"n't ","nt ");
   	std::string phonemized_prompt = phmzr->text_to_phonemes(prompt);
 	// printf("\nRESULT: %s\n",phonemized_prompt.c_str());
 

otherarch/ttscpp/src/phonemizer.cpp

@@ -851,11 +851,15 @@ bool phonemizer::process_word(corpus* text, std::string* output, std::string wor
 	size_t unaccented_size_difference = 0;
 
 	std::string foundstr = found_word_to_ipa(word);
-	if(foundstr!="")
+	if(foundstr!="") //do not use if its part of a contracted word
 	{
-		output->append(foundstr);
-		text->size_pop(word.size());
-		return true;
+		std::string detected = text->next(word.size()+1);
+		if(detected.back()!='\'')
+		{
+			output->append(foundstr);
+			text->size_pop(word.size());
+			return true;
+		}
 	}
 
 	if (has_accent) {