[MIRROR] Adds TTS to the game. Players can select their own voices in preferences. [MDB IGNORE] (#21232)

* Adds TTS to the game. Players can select their own voices in preferences.

* [SEMI-MODULAR] [MIRROR FIX] Fixes the TTS PR. (#21267)

Fixes the TTS PR.

---------

Co-authored-by: Watermelon914 <37270891+Watermelon914@users.noreply.github.com>
Co-authored-by: Iamgoofball <iamgoofball@gmail.com>

code/__DEFINES/dcs/signals/signals_mob/signals_mob_living.dm

@@ -107,7 +107,9 @@
 /// From mob/living/treat_message(): (list/message_args)
 #define COMSIG_LIVING_TREAT_MESSAGE "living_treat_message"
 	/// The index of message_args that corresponds to the actual message
-	#define TREAT_MESSAGE_MESSAGE 1
+	#define TREAT_MESSAGE_ARG 1
+	#define TREAT_TTS_MESSAGE_ARG 2
+	#define TREAT_TTS_FILTER_ARG 3
 
 ///From obj/item/toy/crayon/spraycan
 #define COMSIG_LIVING_MOB_PAINTED "living_mob_painted"

code/__DEFINES/speech_controller.dm

@@ -4,9 +4,3 @@
 #define SPEECH_CONTROLLER_QUEUE_WHISPER_VERB "whisper_verb"
 
 #define SPEECH_CONTROLLER_QUEUE_EMOTE_VERB "emote_verb"
-
-#define MOB_INDEX 1
-
-#define MESSAGE_INDEX 2
-
-#define CATEGORY_INDEX 3

code/__DEFINES/subsystems.dm

@@ -161,6 +161,7 @@
 #define INIT_ORDER_OUTPUTS 35
 #define INIT_ORDER_RESTAURANT 34
 #define INIT_ORDER_POLLUTION 32 //SKYRAT EDIT ADDITION - //Needs to be above atoms
+#define INIT_ORDER_TTS 33
 #define INIT_ORDER_ATOMS 30
 #define INIT_ORDER_ARMAMENTS 27 // SKYRAT EDIT ADDITION - Needs to be between atoms and default so it runs before gun companies
 #define INIT_ORDER_LANGUAGE 25
@@ -225,6 +226,7 @@
 #define FIRE_PRIORITY_STATPANEL 390
 #define FIRE_PRIORITY_CHAT 400
 #define FIRE_PRIORITY_RUNECHAT 410
+#define FIRE_PRIORITY_TTS 425
 #define FIRE_PRIORITY_MOUSE_ENTERED 450
 #define FIRE_PRIORITY_OVERLAYS 500
 #define FIRE_PRIORITY_EXPLOSIONS 666

code/__HELPERS/heap.dm

@@ -20,7 +20,7 @@
 	return !length(L)
 
 //insert and place at its position a new node in the heap
-/datum/heap/proc/insert(atom/A)
+/datum/heap/proc/insert(A)
 
 	L.Add(A)
 	swim(length(L))
@@ -70,7 +70,7 @@
 		return index * 2
 
 //Replaces a given node so it verify the heap condition
-/datum/heap/proc/resort(atom/A)
+/datum/heap/proc/resort(A)
 	var/index = L.Find(A)
 
 	swim(index)

code/__HELPERS/tts.dm

@@ -0,0 +1,4 @@
+/proc/tts_speech_filter(text)
+	// Only allow alphanumeric characters and whitespace
+	var/static/regex/bad_chars_regex = regex("\[^a-zA-Z0-9 ,?.!'&-]", "g")
+	return bad_chars_regex.Replace(text, " ")

code/controllers/configuration/entries/game_options.dm

@@ -418,4 +418,14 @@
 
 /datum/config_entry/flag/disallow_circuit_sounds
 
+/datum/config_entry/string/tts_http_url
+	protection = CONFIG_ENTRY_LOCKED
+
+/datum/config_entry/string/tts_http_token
+	protection = CONFIG_ENTRY_LOCKED|CONFIG_ENTRY_HIDDEN
+
+/datum/config_entry/number/tts_max_concurrent_requests
+	default = 4
+	min_val = 1
+
 /datum/config_entry/flag/give_tutorials_without_db

code/controllers/subsystem/communications.dm

@@ -36,7 +36,8 @@ SUBSYSTEM_DEF(communications)
 		minor_announce(html_decode(input),"[user.name] Announces:", players = players)
 		COOLDOWN_START(src, silicon_message_cooldown, COMMUNICATION_COOLDOWN_AI)
 	else
-		priority_announce(html_decode(user.treat_message(input)), null, ANNOUNCER_CAPTAIN, "[syndicate? "Syndicate " : ""]Captain", has_important_message = TRUE, players = players)//SKYRAT EDIT CHANGE
+		var/list/message_data = user.treat_message(input)
+		priority_announce(html_decode(message_data["message"]), null, ANNOUNCER_CAPTAIN, "[syndicate? "Syndicate " : ""]Captain", has_important_message = TRUE, players = players)
 		COOLDOWN_START(src, nonsilicon_message_cooldown, COMMUNICATION_COOLDOWN)
 	user.log_talk(input, LOG_SAY, tag="priority announcement")
 	message_admins("[ADMIN_LOOKUPFLW(user)] has made a priority announcement.")

code/controllers/subsystem/tts.dm

@@ -0,0 +1,228 @@
+#define TARGET_INDEX 1
+#define IDENTIFIER_INDEX 2
+#define START_TIME_INDEX 3
+#define REQUEST_INDEX 4
+#define MESSAGE_INDEX 5
+
+SUBSYSTEM_DEF(tts)
+	name = "Text To Speech"
+	wait = 0.05 SECONDS
+	priority = FIRE_PRIORITY_TTS
+	init_order = INIT_ORDER_TTS
+	runlevels = RUNLEVEL_LOBBY | RUNLEVEL_SETUP | RUNLEVEL_GAME | RUNLEVEL_POSTGAME
+
+	/// Queued HTTP requests that have yet to be sent. TTS requests are handled as lists rather than datums.
+	/// It could be worth refactoring TTS messages to be datums instead to reduce complexity.
+	var/datum/heap/queued_tts_messages
+
+	/// HTTP requests currently in progress but not being processed yet
+	var/list/in_process_tts_messages = list()
+
+	/// HTTP requests that are being processed to see if they've been finished
+	var/list/current_processing_tts_messages = list()
+
+	/// A list of available speakers, which are string identifiers of the TTS voices that can be used to generate TTS messages.
+	var/list/available_speakers = list()
+
+	/// A list of current tts messages being processed, mapped by their sha1 identifier.
+	/// Used to prevent double processing of the same message, voice and filter, since we can just
+	/// cache extra requests to the current tts message being processed at once and play them upon request completion.
+	var/list/cached_voices = list()
+
+	/// Whether TTS is enabled or not
+	var/tts_enabled = FALSE
+
+	/// TTS messages won't play if requests took longer than this duration of time.
+	var/message_timeout = 7 SECONDS
+
+	/// Messages can be timed out earlier if the algorithm thinks that
+	/// it's going to take too long for their message to be processed.
+	/// This'll determine the minimum extent of how late it is allowed to begin timing messages out
+	var/message_timeout_early_minimum = 5 SECONDS
+
+	/// The max concurrent http requests that can be made at one time. Used to prevent 1 server from overloading the tts server
+	var/max_concurrent_requests = 4
+
+	/// Used to calculate the average time it takes for a tts message to be received from the http server
+	/// For tts messages which time out, it won't keep tracking the tts message and will just assume that the message took
+	/// 7 seconds (or whatever the value of message_timeout is) to receive back a response.
+	var/average_tts_messages_time = 0
+
+/datum/controller/subsystem/tts/vv_edit_var(var_name, var_value)
+	// tts being enabled depends on whether it actually exists
+	if(NAMEOF(src, tts_enabled) == var_name)
+		return FALSE
+	return ..()
+
+/datum/controller/subsystem/tts/stat_entry(msg)
+	msg = "Active:[length(in_process_tts_messages)]|Standby:[length(queued_tts_messages.L)]|Avg:[average_tts_messages_time]"
+	return ..()
+
+/proc/cmp_word_length_asc(list/a, list/b)
+	return length(b[MESSAGE_INDEX]) - length(a[MESSAGE_INDEX])
+
+/datum/controller/subsystem/tts/Initialize()
+	if(!CONFIG_GET(string/tts_http_url))
+		return SS_INIT_NO_NEED
+
+	queued_tts_messages = new /datum/heap(GLOBAL_PROC_REF(cmp_word_length_asc))
+	var/datum/http_request/request = new()
+	var/list/headers = list()
+	headers["Authorization"] = CONFIG_GET(string/tts_http_token)
+	request.prepare(RUSTG_HTTP_METHOD_GET, "[CONFIG_GET(string/tts_http_url)]/tts-voices", "", headers)
+	request.begin_async()
+	UNTIL(request.is_complete())
+	var/datum/http_response/response = request.into_response()
+	if(response.errored || response.status_code != 200)
+		stack_trace(response.error)
+		return SS_INIT_FAILURE
+	max_concurrent_requests = CONFIG_GET(number/tts_max_concurrent_requests)
+	available_speakers = json_decode(response.body)
+	tts_enabled = TRUE
+	rustg_file_write(json_encode(available_speakers), "data/cached_tts_voices.json")
+	rustg_file_write("rustg HTTP requests can't write to folders that don't exist, so we need to make it exist.", "tmp/tts/init.txt")
+	return SS_INIT_SUCCESS
+
+/datum/controller/subsystem/tts/proc/play_tts(target, sound/audio, datum/language/language, local, range = 7)
+	if(local)
+		SEND_SOUND(target, audio)
+		return
+
+	var/turf/turf_source = get_turf(target)
+	if(!turf_source)
+		return
+
+	var/channel = SSsounds.random_available_channel()
+	var/listeners = get_hearers_in_view(range, turf_source)
+
+	for(var/mob/listening_mob in listeners | SSmobs.dead_players_by_zlevel[turf_source.z])//observers always hear through walls
+		var/datum/language_holder/holder = listening_mob.get_language_holder()
+		if(!listening_mob.client?.prefs.read_preference(/datum/preference/toggle/sound_tts))
+			continue
+
+		if(get_dist(listening_mob, turf_source) <= range && holder.has_language(language, spoken = FALSE))
+			listening_mob.playsound_local(
+				turf_source,
+				vol = (listening_mob == target)? 60 : 85,
+				falloff_exponent = SOUND_FALLOFF_EXPONENT,
+				channel = channel,
+				pressure_affected = TRUE,
+				sound_to_use = audio,
+				max_distance = SOUND_RANGE,
+				falloff_distance = SOUND_DEFAULT_FALLOFF_DISTANCE,
+				distance_multiplier = 1,
+				use_reverb = TRUE
+			)
+
+/datum/controller/subsystem/tts/proc/handle_request(list/entry)
+	var/timeout_time = entry[START_TIME_INDEX] + message_timeout
+	if(timeout_time < world.time)
+		cached_voices -= entry[IDENTIFIER_INDEX]
+		return
+	var/datum/http_request/request = entry[REQUEST_INDEX]
+	request.begin_async()
+	in_process_tts_messages += list(entry)
+
+// Need to wait for all HTTP requests to complete here because of a rustg crash bug that causes crashes when dd restarts whilst HTTP requests are ongoing.
+/datum/controller/subsystem/tts/Shutdown()
+	tts_enabled = FALSE
+	for(var/list/data in in_process_tts_messages)
+		var/datum/http_request/request = data[REQUEST_INDEX]
+		UNTIL(request.is_complete())
+
+/datum/controller/subsystem/tts/fire(resumed)
+	if(!tts_enabled)
+		flags |= SS_NO_FIRE
+		return
+
+	if(!resumed)
+		while(length(in_process_tts_messages) < max_concurrent_requests && length(queued_tts_messages.L) > 0)
+			var/list/entry = queued_tts_messages.pop()
+			handle_request(entry)
+		current_processing_tts_messages = in_process_tts_messages.Copy()
+
+	// For speed
+	var/list/processing_messages = current_processing_tts_messages
+	while(processing_messages.len)
+		var/current_message = processing_messages[processing_messages.len]
+		processing_messages.len--
+		var/datum/http_request/request = current_message[REQUEST_INDEX]
+		if(!request.is_complete())
+			continue
+
+		var/datum/http_response/response = request.into_response()
+		in_process_tts_messages -= list(current_message)
+		average_tts_messages_time = MC_AVERAGE(average_tts_messages_time, world.time - current_message[START_TIME_INDEX])
+		// If it took too long to process, don't bother playing it
+		var/timeout_time = current_message[START_TIME_INDEX] + message_timeout
+		var/identifier = current_message[IDENTIFIER_INDEX]
+		cached_voices -= identifier
+		if(response.errored || timeout_time < world.time)
+			continue
+
+		var/sound/new_sound = new("tmp/tts/[identifier].ogg")
+		for(var/target in current_message[TARGET_INDEX])
+			play_tts(target["target"], new_sound, target["language"], target["local"], target["range"])
+		if(MC_TICK_CHECK)
+			return
+
+#define ADD_TARGET_TO_STRUCT(tts_struct, target, language, local, range) ##tts_struct[TARGET_INDEX] += list(list("target" = ##target, "language" = ##language, "local" = ##local, "range" = ##range))
+
+/datum/controller/subsystem/tts/proc/queue_tts_message(target, message, datum/language/language, speaker, filter, local = FALSE, message_range = 7)
+	if(!tts_enabled)
+		return
+
+	var/static/regex/contains_alphanumeric = regex("\[a-zA-Z0-9]")
+	// If there is no alphanumeric char, the output will usually be static, so
+	// don't bother sending
+	if(contains_alphanumeric.Find(message) == 0)
+		return
+
+	var/shell_scrubbed_input = tts_speech_filter(message)
+	shell_scrubbed_input = copytext(shell_scrubbed_input, 1, 300)
+	var/identifier = sha1(speaker + filter + shell_scrubbed_input)
+	var/cached_voice = cached_voices[identifier]
+	if(islist(cached_voice))
+		ADD_TARGET_TO_STRUCT(cached_voice, target, language, local, message_range)
+		return
+	else if(fexists("tmp/tts/[identifier].ogg"))
+		var/sound/new_sound = new("tmp/tts/[identifier].ogg")
+		play_tts(target, new_sound, language, local, message_range)
+		return
+	if(!(speaker in available_speakers))
+		return
+
+	var/list/headers = list()
+	headers["Content-Type"] = "application/json"
+	headers["Authorization"] = CONFIG_GET(string/tts_http_token)
+	var/datum/http_request/request = new()
+	var/file_name = "tmp/tts/[identifier].ogg"
+	request.prepare(RUSTG_HTTP_METHOD_GET, "[CONFIG_GET(string/tts_http_url)]/tts?voice=[speaker]&identifier=[identifier]&filter=[url_encode(filter)]", json_encode(list("text" = shell_scrubbed_input)), headers, file_name)
+	// This'll probably be better off datumized in the future, but it's not necessary to do right now
+	var/list/data = list(
+		// TARGET_INDEX = 1
+		list(),
+		// IDENTIFIER_INDEX = 2
+		identifier,
+		// START_TIME_INDEX = 3
+		world.time,
+		// REQUEST_INDEX = 4
+		request,
+		// MESSAGE_INDEX = 5
+		shell_scrubbed_input,
+	)
+	ADD_TARGET_TO_STRUCT(data, target, language, local, message_range)
+	cached_voices[identifier] = data
+	if(length(in_process_tts_messages) < max_concurrent_requests)
+		request.begin_async()
+		in_process_tts_messages += list(data)
+	else
+		queued_tts_messages.insert(list(data))
+
+#undef ADD_TARGET_TO_STRUCT
+
+#undef TARGET_INDEX
+#undef IDENTIFIER_INDEX
+#undef START_TIME_INDEX
+#undef REQUEST_INDEX
+#undef MESSAGE_INDEX

code/datums/components/sign_language.dm

@@ -227,7 +227,9 @@
 	SIGNAL_HANDLER
 
 	if(check_signables_state() == SIGN_ONE_HAND)
-		message_args[TREAT_MESSAGE_MESSAGE] = stars(message_args[TREAT_MESSAGE_MESSAGE])
+		message_args[TREAT_MESSAGE_ARG] = stars(message_args[TREAT_MESSAGE_ARG])
+
+	message_args[TREAT_TTS_MESSAGE_ARG] = ""
 
 /// Signal proc for [COMSIG_MOVABLE_SAY_QUOTE]
 /// Removes exclamation/question marks.

code/datums/quirks/negative_quirks.dm

@@ -694,8 +694,8 @@
 			if(prob(max(5,(nearby_people*12.5*moodmod)))) //Minimum 1/20 chance of stutter
 				// Add a short stutter, THEN treat our word
 				quirker.adjust_stutter(0.5 SECONDS)
-				new_message += quirker.treat_message(word, capitalize_message = FALSE)
-
+				var/list/message_data = quirker.treat_message(word, capitalize_message = FALSE)
+				new_message += message_data["message"]
 			else
 				new_message += word
 

code/datums/status_effects/debuffs/speech_debuffs.dm

@@ -3,6 +3,9 @@
 	alert_type = null
 	remove_on_fullheal = TRUE
 
+	var/make_tts_message_original = FALSE
+	var/tts_filter = ""
+
 /datum/status_effect/speech/on_creation(mob/living/new_owner, duration = 10 SECONDS)
 	src.duration = duration
 	return ..()
@@ -23,10 +26,15 @@
 /datum/status_effect/speech/proc/handle_message(datum/source, list/message_args)
 	SIGNAL_HANDLER
 
-	var/phrase = html_decode(message_args[TREAT_MESSAGE_MESSAGE])
+	var/phrase = html_decode(message_args[TREAT_MESSAGE_ARG])
 	if(!length(phrase))
 		return
 
+	if(length(tts_filter) > 0)
+		message_args[TREAT_TTS_FILTER_ARG] += tts_filter
+	if(make_tts_message_original)
+		message_args[TREAT_TTS_MESSAGE_ARG] = message_args[TREAT_MESSAGE_ARG]
+
 	var/final_phrase = ""
 	var/original_char = ""
 
@@ -35,7 +43,7 @@
 
 		final_phrase += apply_speech(original_char, original_char)
 
-	message_args[TREAT_MESSAGE_MESSAGE] = sanitize(final_phrase)
+	message_args[TREAT_MESSAGE_ARG] = sanitize(final_phrase)
 
 /**
  * Applies the speech effects on the past character, changing
@@ -54,6 +62,9 @@
 	/// Regex of characters we won't apply a stutter to
 	var/static/regex/no_stutter
 
+	make_tts_message_original = TRUE
+	tts_filter = "tremolo=f=10:d=0.8,rubberband=tempo=0.5"
+
 /datum/status_effect/speech/stutter/on_creation(mob/living/new_owner, ...)
 	. = ..()
 	if(!.)
@@ -83,7 +94,7 @@
 
 /datum/status_effect/speech/stutter/derpspeech/handle_message(datum/source, list/message_args)
 
-	var/message = html_decode(message_args[TREAT_MESSAGE_MESSAGE])
+	var/message = html_decode(message_args[TREAT_MESSAGE_ARG])
 
 	message = replacetext(message, " am ", " ")
 	message = replacetext(message, " is ", " ")
@@ -100,7 +111,7 @@
 		message = uppertext(message)
 		message += "[apply_speech(exclamation, exclamation)]"
 
-	message_args[1] = message
+	message_args[TREAT_MESSAGE_ARG] = message
 
 	var/mob/living/living_source = source
 	if(!isliving(source) || living_source.has_status_effect(/datum/status_effect/speech/stutter))
@@ -229,6 +240,8 @@
 	doubletext_prob = 0
 	text_modification_file = "slurring_cult_text.json"
 
+	tts_filter = "rubberband=pitch=0.5,vibrato=5"
+
 /datum/status_effect/speech/slurring/heretic
 	id = "heretic_slurring"
 	common_prob = 50

code/game/atoms_movable.dm

@@ -96,6 +96,13 @@
 
 	/// Whether a user will face atoms on entering them with a mouse. Despite being a mob variable, it is here for performances //SKYRAT EDIT ADDITION
 	var/face_mouse = FALSE //SKYRAT EDIT ADDITION
+
+	/// The voice that this movable makes when speaking
+	var/voice
+
+	/// The filter to apply to the voice when processing the TTS audio message.
+	var/voice_filter = ""
+
 	/// Value used to increment ex_act() if reactionary_explosions is on
 	/// How much we as a source block explosions by
 	/// Will not automatically apply to the turf below you, you need to apply /datum/element/block_explosives in conjunction with this

code/game/machinery/computer/communications.dm

@@ -778,7 +778,8 @@ GLOBAL_VAR_INIT(cops_arrived, FALSE)
 		return
 	if(user.try_speak(input))
 		//Adds slurs and so on. Someone should make this use languages too.
-		input = user.treat_message(input)
+		var/list/input_data = user.treat_message(input)
+		input = input_data["message"]
 	else
 		//No cheating, mime/random mute guy!
 		input = "..."

code/game/say.dm

@@ -80,12 +80,29 @@ GLOBAL_LIST_INIT(freqtospan, list(
 /atom/movable/proc/can_speak(allow_mimes = FALSE)
 	return TRUE
 
-/atom/movable/proc/send_speech(message, range = 7, obj/source = src, bubble_type, list/spans, datum/language/message_language, list/message_mods = list(), forced = FALSE)
+/atom/movable/proc/send_speech(message, range = 7, obj/source = src, bubble_type, list/spans, datum/language/message_language, list/message_mods = list(), forced = FALSE, tts_message, list/tts_filter)
+	var/found_client = FALSE
 	for(var/atom/movable/hearing_movable as anything in get_hearers_in_view(range, source))
 		if(!hearing_movable)//theoretically this should use as anything because it shouldnt be able to get nulls but there are reports that it does.
 			stack_trace("somehow theres a null returned from get_hearers_in_view() in send_speech!")
 			continue
 		hearing_movable.Hear(null, src, message_language, message, null, spans, message_mods, range)
+		if(!found_client && length(hearing_movable.client_mobs_in_contents))
+			found_client = TRUE
+
+	var/tts_message_to_use = tts_message
+	if(!tts_message_to_use)
+		tts_message_to_use = message
+
+	var/list/filter = list()
+	if(length(voice_filter) > 0)
+		filter += voice_filter
+
+	if(length(tts_filter) > 0)
+		filter += tts_filter.Join(",")
+
+	if(voice && found_client)
+		INVOKE_ASYNC(SStts, TYPE_PROC_REF(/datum/controller/subsystem/tts, queue_tts_message), src, html_decode(tts_message_to_use), message_language, voice, filter.Join(","), message_range = range)
 
 /atom/movable/proc/compose_message(atom/movable/speaker, datum/language/message_language, raw_message, radio_freq, list/spans, list/message_mods = list(), face_name = FALSE)
 	//This proc uses text() because it is faster than appending strings. Thanks BYOND.

code/modules/antagonists/changeling/changeling.dm

@@ -605,6 +605,8 @@
 		new_profile.worn_icon_vox_list[slot] = clothing_item.worn_icon_vox
 		new_profile.supports_variations_flags_list[slot] = clothing_item.supports_variations_flags
 		// SKYRAT EDIT END
+	new_profile.voice = target.voice
+	new_profile.voice_filter = target.voice_filter
 
 	return new_profile
 
@@ -799,6 +801,7 @@
 	user.dna.body_markings = chosen_dna.body_markings.Copy()
 	user.grad_style = LAZYLISTDUPLICATE(chosen_profile.grad_style)
 	user.grad_color = LAZYLISTDUPLICATE(chosen_profile.grad_color)
+
 	user.physique = chosen_profile.physique
 	qdel(user.selected_scream)
 	qdel(user.selected_laugh)
@@ -821,6 +824,8 @@
 				break
 
 	// SKYRAT EDIT END
+	user.voice = chosen_profile.voice
+	user.voice_filter = chosen_profile.voice_filter
 
 	chosen_dna.transfer_identity(user, TRUE)
 
@@ -982,6 +987,7 @@
 	var/emissive_eyes
 	var/list/grad_style = list("None", "None")
 	var/list/grad_color = list(null, null)
+
 	var/physique
 	var/list/worn_icon_digi_list = list()
 	var/list/worn_icon_monkey_list = list()
@@ -994,6 +1000,12 @@
 	var/list/quirks = list()
 	/// SKYRAT EDIT END
 
+	/// The TTS voice of the profile source
+	var/voice
+	/// The TTS filter of the profile filter
+	var/voice_filter = ""
+
+
 /datum/changeling_profile/Destroy()
 	qdel(dna)
 	LAZYCLEARLIST(stored_scars)
@@ -1034,6 +1046,7 @@
 	new_profile.emissive_eyes = emissive_eyes
 	new_profile.grad_style = LAZYLISTDUPLICATE(grad_style)
 	new_profile.grad_color = LAZYLISTDUPLICATE(grad_color)
+
 	new_profile.physique = physique
 	new_profile.worn_icon_digi_list = worn_icon_digi_list.Copy()
 	new_profile.worn_icon_monkey_list = worn_icon_monkey_list.Copy()
@@ -1046,6 +1059,9 @@
 	new_profile.quirks = quirks.Copy()
 	// SKYRAT EDIT END
 
+	new_profile.voice = voice
+	new_profile.voice_filter = voice_filter
+
 /datum/antagonist/changeling/roundend_report()
 	var/list/parts = list()
 

code/modules/client/preferences/middleware/tts.dm

@@ -0,0 +1,16 @@
+/// Middleware to handle quirks
+/datum/preference_middleware/tts
+	/// Cooldown on requesting a TTS preview.
+	COOLDOWN_DECLARE(tts_test_cooldown)
+
+	action_delegations = list(
+		"play_voice" = PROC_REF(play_voice),
+	)
+
+/datum/preference_middleware/tts/proc/play_voice(list/params, mob/user)
+	if(!COOLDOWN_FINISHED(src, tts_test_cooldown))
+		return TRUE
+	var/speaker = preferences.read_preference(/datum/preference/choiced/voice)
+	COOLDOWN_START(src, tts_test_cooldown, 0.5 SECONDS)
+	INVOKE_ASYNC(SStts, TYPE_PROC_REF(/datum/controller/subsystem/tts, queue_tts_message), user.client, "Hello, this is my voice.", speaker = speaker, local = TRUE)
+	return TRUE

code/modules/client/preferences/sounds.dm

@@ -28,6 +28,11 @@
 	savefile_key = "sound_instruments"
 	savefile_identifier = PREFERENCE_PLAYER
 
+/datum/preference/toggle/sound_tts
+	category = PREFERENCE_CATEGORY_GAME_PREFERENCES
+	savefile_key = "sound_tts"
+	savefile_identifier = PREFERENCE_PLAYER
+
 /// Controls hearing dance machines
 /datum/preference/toggle/sound_jukebox
 	category = PREFERENCE_CATEGORY_GAME_PREFERENCES

code/modules/client/preferences/voice.dm

@@ -0,0 +1,26 @@
+/// TTS voice preference
+/datum/preference/choiced/voice
+	savefile_identifier = PREFERENCE_CHARACTER
+	savefile_key = "tts_voice"
+	category = PREFERENCE_CATEGORY_NON_CONTEXTUAL
+
+/datum/preference/choiced/voice/is_accessible(datum/preferences/preferences)
+	if(!SStts.tts_enabled)
+		return FALSE
+	return ..()
+
+/datum/preference/choiced/voice/init_possible_values()
+	if(SStts.tts_enabled)
+		return SStts.available_speakers
+	if(fexists("data/cached_tts_voices.json"))
+		var/list/text_data = rustg_file_read("data/cached_tts_voices.json")
+		var/list/cached_data = json_decode(text_data)
+		if(!cached_data)
+			return list("invalid")
+		return cached_data
+	return list("invalid")
+
+/datum/preference/choiced/voice/apply_to_human(mob/living/carbon/human/target, value)
+	if(SStts.tts_enabled && !(value in SStts.available_speakers))
+		value = pick(SStts.available_speakers) // As a failsafe
+	target.voice = value

code/modules/mob/living/brain/brain_say.dm

@@ -19,7 +19,7 @@
 	else
 		return ..()
 
-/mob/living/brain/treat_message(message, capitalize_message = TRUE)
+/mob/living/brain/treat_message(message, tts_message, tts_filter, capitalize_message = TRUE)
 	if(capitalize_message)
 		message = capitalize(message)
-	return message
+	return list(message = message, tts_message = message, tts_filter = list())

code/modules/mob/living/carbon/human/login.dm

@@ -3,6 +3,9 @@
 
 	dna?.species?.on_owner_login(src)
 
+	if(SStts.tts_enabled && !voice)
+		voice = pick(SStts.available_speakers)
+
 	if(!LAZYLEN(afk_thefts))
 		return
 

code/modules/mob/living/living_say.dm

@@ -184,7 +184,10 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
 		else
 			log_talk(message, LOG_SAY, forced_by = forced, custom_say_emote = message_mods[MODE_CUSTOM_SAY_EMOTE])
 
-	message = treat_message(message) // unfortunately we still need this
+	var/list/message_data = treat_message(message) // unfortunately we still need this
+	message = message_data["message"]
+	var/tts_message = message_data["tts_message"]
+	var/list/tts_filter = message_data["tts_filter"]
 
 	spans |= speech_span
 
@@ -205,10 +208,14 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
 
 	if(!HAS_TRAIT(src, TRAIT_SIGN_LANG)) // if using sign language skip sending the say signal
 		// Make sure the arglist is passed exactly - don't pass a copy of it. Say signal handlers will modify some of the parameters.
+		var/last_message = message
 		var/sigreturn = SEND_SIGNAL(src, COMSIG_MOB_SAY, args)
+		if(last_message != message)
+			tts_message = message
 		if(sigreturn & COMPONENT_UPPERCASE_SPEECH)
 			message = uppertext(message)
 
+
 	if(!message)
 		if(succumbed)
 			succumb()
@@ -243,8 +250,7 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
 	if(pressure < ONE_ATMOSPHERE*0.4) //Thin air, let's italicise the message
 		spans |= SPAN_ITALICS
 
-	send_speech(message, message_range, src, bubble_type, spans, language, message_mods)//roughly 58% of living/say()'s total cost
-
+	send_speech(message, message_range, src, bubble_type, spans, language, message_mods, tts_message = tts_message, tts_filter = tts_filter)//roughly 58% of living/say()'s total cost
 	if(succumbed)
 		succumb(TRUE)
 		to_chat(src, compose_message(src, language, message, , spans, message_mods))
@@ -326,7 +332,7 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
 	show_message(message, MSG_AUDIBLE, deaf_message, deaf_type, avoid_highlight)
 	return message
 
-/mob/living/send_speech(message_raw, message_range = 6, obj/source = src, bubble_type = bubble_icon, list/spans, datum/language/message_language = null, list/message_mods = list(), forced = null)
+/mob/living/send_speech(message_raw, message_range = 6, obj/source = src, bubble_type = bubble_icon, list/spans, datum/language/message_language = null, list/message_mods = list(), forced = null, tts_message, list/tts_filter)
 	var/whisper_range = 0
 	var/is_speaker_whispering = FALSE
 	if(message_mods[WHISPER_MODE]) //If we're whispering
@@ -365,10 +371,27 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
 
 	//speech bubble
 	var/list/speech_bubble_recipients = list()
+	var/found_client = FALSE
 	var/talk_icon_state = say_test(message_raw)
 	for(var/mob/M in listening)
-		if(M.client && (!M.client.prefs.read_preference(/datum/preference/toggle/enable_runechat) || (SSlag_switch.measures[DISABLE_RUNECHAT] && !HAS_TRAIT(src, TRAIT_BYPASS_MEASURES))))
-			speech_bubble_recipients.Add(M.client)
+		if(M.client)
+			if(!M.client.prefs.read_preference(/datum/preference/toggle/enable_runechat) || (SSlag_switch.measures[DISABLE_RUNECHAT] && !HAS_TRAIT(src, TRAIT_BYPASS_MEASURES)))
+				speech_bubble_recipients.Add(M.client)
+			found_client = TRUE
+
+	if(voice && found_client && !message_mods[MODE_CUSTOM_SAY_ERASE_INPUT] && !HAS_TRAIT(src, TRAIT_SIGN_LANG))
+		var/tts_message_to_use = tts_message
+		if(!tts_message_to_use)
+			tts_message_to_use = message_raw
+
+		var/list/filter = list()
+		if(length(voice_filter) > 0)
+			filter += voice_filter
+
+		if(length(tts_filter) > 0)
+			filter += tts_filter.Join(",")
+
+		INVOKE_ASYNC(SStts, TYPE_PROC_REF(/datum/controller/subsystem/tts, queue_tts_message), src, html_decode(tts_message_to_use), message_language, voice, filter.Join(","), message_range = message_range)
 
 	var/image/say_popup = image('icons/mob/effects/talk.dmi', src, "[bubble_type][talk_icon_state]", FLY_LAYER)
 	SET_PLANE_EXPLICIT(say_popup, ABOVE_GAME_PLANE, src)
@@ -421,16 +444,25 @@ GLOBAL_LIST_INIT(message_modes_stat_limits, list(
  * message - The message to treat.
  * capitalize_message - Whether we run capitalize() on the message after we're done.
  */
-/mob/living/proc/treat_message(message, capitalize_message = TRUE)
+/mob/living/proc/treat_message(message, tts_message, tts_filter, capitalize_message = TRUE)
 	if(HAS_TRAIT(src, TRAIT_UNINTELLIGIBLE_SPEECH))
 		message = unintelligize(message)
 
-	SEND_SIGNAL(src, COMSIG_LIVING_TREAT_MESSAGE, args)
+	tts_filter = list()
+	var/list/data = list(message, tts_message, tts_filter)
+	SEND_SIGNAL(src, COMSIG_LIVING_TREAT_MESSAGE, data)
+	message = data[TREAT_MESSAGE_ARG]
+	tts_message = data[TREAT_TTS_MESSAGE_ARG]
+	tts_filter = data[TREAT_TTS_FILTER_ARG]
+
+	if(!tts_message)
+		tts_message = message
 
 	if(capitalize_message)
 		message = capitalize(message)
+		tts_message = capitalize(tts_message)
 
-	return message
+	return list(message = message, tts_message = tts_message, tts_filter = tts_filter)
 
 /mob/living/proc/radio(message, list/message_mods = list(), list/spans, language)
 	//SKYRAT EDIT ADDITION BEGIN

code/modules/mob/living/silicon/login.dm

@@ -1,6 +1,12 @@
 /mob/living/silicon/Login()
 	if(mind)
 		mind?.remove_antags_for_borging()
+	// SKYRAT EDIT BEGIN - Let people set a TTS voice for their silicon characters and pAIs
+	if(SStts.tts_enabled)
+		var/voice_to_use = client?.prefs.read_preference(/datum/preference/choiced/voice)
+		if(voice_to_use)
+			voice = voice_to_use
+	// SKYRAT EDIT END
 	return ..()
 
 

code/modules/mob/living/silicon/silicon.dm

@@ -13,6 +13,7 @@
 	flags_1 = PREVENT_CONTENTS_EXPLOSION_1
 	examine_cursor_icon = null
 	fire_stack_decay_rate = -0.55
+	voice_filter = "afftfilt=real='hypot(re,im)*sin(0)':imag='hypot(re,im)*cos(0)':win_size=512:overlap=1,rubberband=pitch=0.8"
 	var/datum/ai_laws/laws = null//Now... THEY ALL CAN ALL HAVE LAWS
 	var/last_lawchange_announce = 0
 	var/list/alarms_to_show = list()
@@ -51,8 +52,11 @@
 
 	var/obj/item/modular_computer/pda/silicon/modularInterface
 
+
 /mob/living/silicon/Initialize(mapload)
 	. = ..()
+	if(SStts.tts_enabled)
+		voice = pick(SStts.available_speakers)
 	GLOB.silicon_mobs += src
 	faction += FACTION_SILICON
 	if(ispath(radio))

code/modules/mob/living/simple_animal/friendly/robot_customer.dm

@@ -82,7 +82,7 @@
 	if(bonus_overlays)
 		. += bonus_overlays
 
-/mob/living/simple_animal/robot_customer/send_speech(message, message_range, obj/source, bubble_type, list/spans, datum/language/message_language, list/message_mods, forced)
+/mob/living/simple_animal/robot_customer/send_speech(message, message_range, obj/source, bubble_type, list/spans, datum/language/message_language, list/message_mods, forced, tts_message, list/tts_filter)
 	. = ..()
 	var/datum/customer_data/customer_info = ai_controller.blackboard[BB_CUSTOMER_CUSTOMERINFO]
 	playsound(src, customer_info.speech_sound, 30, extrarange = MEDIUM_RANGE_SOUND_EXTRARANGE, falloff_distance = 5)

code/modules/mob/living/simple_animal/parrot.dm

@@ -901,13 +901,18 @@ GLOBAL_LIST_INIT(strippable_parrot_items, create_strippable_list(list(
 	speak = list("Poly wanna cracker!", ":e Check the crystal, you chucklefucks!",":e Wire the solars, you lazy bums!",":e WHO TOOK THE DAMN MODSUITS?",":e OH GOD ITS ABOUT TO DELAMINATE CALL THE SHUTTLE")
 	gold_core_spawnable = NO_SPAWN
 	speak_chance = 3
+	voice_filter = "rubberband=pitch=1.5"
+
 	var/memory_saved = FALSE
 	var/rounds_survived = 0
 	var/longest_survival = 0
 	var/longest_deathstreak = 0
 
+
 /mob/living/simple_animal/parrot/poly/Initialize(mapload)
 	ears = new /obj/item/radio/headset/headset_eng(src)
+	if(SStts.tts_enabled)
+		voice = pick(SStts.available_speakers)
 	available_channels = list(":e")
 	Read_Memory()
 	if(rounds_survived == longest_survival)

code/modules/mob/login.dm

@@ -29,6 +29,7 @@
 /mob/Login()
 	if(!client)
 		return FALSE
+
 	canon_client = client
 	add_to_player_list()
 	lastKnownIP = client.address

code/modules/vending/_vending.dm

@@ -64,6 +64,7 @@
 	payment_department = ACCOUNT_SRV
 	light_power = 0.7
 	light_range = MINIMUM_USEFUL_LIGHT_RANGE
+	voice_filter = "aderivative"
 
 	/// Is the machine active (No sales pitches if off)!
 	var/active = 1
@@ -185,7 +186,6 @@
 	/// used for narcing on underages
 	var/obj/item/radio/sec_radio
 
-
 /**
  * Initialize the vending machine
  *
@@ -208,6 +208,12 @@
 	. = ..()
 	wires = new /datum/wires/vending(src)
 
+	if(SStts.tts_enabled)
+		var/static/vendor_voice_by_type = list()
+		if(!vendor_voice_by_type[type])
+			vendor_voice_by_type[type] = pick(SStts.available_speakers)
+		voice = vendor_voice_by_type[type]
+
 	if(build_inv) //non-constructable vending machine
 		build_inventories()
 

config/config.txt

@@ -665,6 +665,15 @@ PR_ANNOUNCEMENTS_PER_ROUND 5
 ## Uncomment to block granting profiling privileges to users with R_DEBUG, for performance purposes
 #FORBID_ADMIN_PROFILING
 
+## Link to a HTTP server that's been set up on a server. Docker-compose file can be found in tools/tts
+#TTS_HTTP_URL http://localhost:5002
+
+## Token that can be used to prevent misuse of your TTS server that you've set up.
+#TTS_HTTP_TOKEN coolio
+
+## The maximum number of concurrent tts http requests that can be made by the server at once.
+#TTS_MAX_CONCURRENT_REQUESTS 4
+
 ## Comment to disable sending a toast notification on the host server when initializations complete.
 ## Even if this is enabled, a notification will only be sent if there are no clients connected.
 TOAST_NOTIFICATION_ON_INIT

interface/skin.dmf

@@ -86,6 +86,14 @@ window "mainwindow"
 		anchor2 = -1,-1
 		is-visible = false
 		saved-params = ""
+	elem "html_audio_player"
+		type = BROWSER
+		pos = 0,0
+		size = 200x200
+		anchor1 = none
+		anchor2 = none
+		is-visible = false
+		saved-params = ""
 	elem "tooltip"
 		type = BROWSER
 		pos = 0,0

tgstation.dme

@@ -486,6 +486,7 @@
 #include "code\__HELPERS\text.dm"
 #include "code\__HELPERS\time.dm"
 #include "code\__HELPERS\traits.dm"
+#include "code\__HELPERS\tts.dm"
 #include "code\__HELPERS\turfs.dm"
 #include "code\__HELPERS\type2type.dm"
 #include "code\__HELPERS\type_processing.dm"
@@ -695,6 +696,7 @@
 #include "code\controllers\subsystem\timer.dm"
 #include "code\controllers\subsystem\title.dm"
 #include "code\controllers\subsystem\traitor.dm"
+#include "code\controllers\subsystem\tts.dm"
 #include "code\controllers\subsystem\tutorials.dm"
 #include "code\controllers\subsystem\verb_manager.dm"
 #include "code\controllers\subsystem\vis_overlays.dm"
@@ -3148,6 +3150,7 @@
 #include "code\modules\client\preferences\ui_style.dm"
 #include "code\modules\client\preferences\underwear_color.dm"
 #include "code\modules\client\preferences\uplink_location.dm"
+#include "code\modules\client\preferences\voice.dm"
 #include "code\modules\client\preferences\widescreen.dm"
 #include "code\modules\client\preferences\window_flashing.dm"
 #include "code\modules\client\preferences\middleware\_middleware.dm"
@@ -3159,6 +3162,7 @@
 #include "code\modules\client\preferences\middleware\quirks.dm"
 #include "code\modules\client\preferences\middleware\random.dm"
 #include "code\modules\client\preferences\middleware\species.dm"
+#include "code\modules\client\preferences\middleware\tts.dm"
 #include "code\modules\client\preferences\migrations\body_type_migration.dm"
 #include "code\modules\client\preferences\migrations\convert_to_json_savefile.dm"
 #include "code\modules\client\preferences\migrations\legacy_sound_toggles_migration.dm"

tgui/packages/tgui/interfaces/PreferencesMenu/preferences/features/character_preferences/tts_voice.tsx

@@ -0,0 +1,29 @@
+import { FeatureChoiced, FeatureChoicedServerData, FeatureDropdownInput, FeatureValueProps } from '../base';
+import { Stack, Button } from '../../../../../components';
+
+const FeatureTTSDropdownInput = (
+  props: FeatureValueProps<string, string, FeatureChoicedServerData>
+) => {
+  return (
+    <Stack>
+      <Stack.Item grow>
+        <FeatureDropdownInput {...props} />
+      </Stack.Item>
+      <Stack.Item>
+        <Button
+          onClick={() => {
+            props.act('play_voice');
+          }}
+          icon="play"
+          width="100%"
+          height="100%"
+        />
+      </Stack.Item>
+    </Stack>
+  );
+};
+
+export const tts_voice: FeatureChoiced = {
+  name: 'Voice',
+  component: FeatureTTSDropdownInput,
+};

tgui/packages/tgui/interfaces/PreferencesMenu/preferences/features/game_preferences/sounds.tsx

@@ -34,6 +34,13 @@ export const sound_instruments: FeatureToggle = {
   component: CheckboxInput,
 };
 
+export const sound_tts: FeatureToggle = {
+  name: 'Enable TTS',
+  category: 'SOUND',
+  description: 'When enabled, be able to hear text-to-speech sounds in game.',
+  component: CheckboxInput,
+};
+
 export const sound_jukebox: FeatureToggle = {
   name: 'Enable jukebox music',
   category: 'SOUND',

tools/tts/docker-compose.yml

@@ -0,0 +1,29 @@
+services:
+  tts-api:
+    container_name: tts-api
+
+    build: ./tts-api
+    ports:
+      - "5002:5002"
+    restart: always
+    healthcheck:
+      test: ["CMD", "curl", "-sf", "http://localhost:5002/health-check"]
+      interval: 1s
+      timeout: 1s
+      retries: 3
+      start_period: 30s
+    links:
+      - "tts:tts-container"
+  tts:
+    container_name: tts
+
+    build: ./tts
+    restart: always
+    healthcheck:
+      test: ["CMD", "curl", "-sf", "http://localhost:5003/health-check"]
+      interval: 1s
+      timeout: 1s
+      retries: 3
+      start_period: 30s
+    volumes:
+      - ./tts/persistent_data/tts_models:/root/tts_data

tools/tts/documentation.md

@@ -0,0 +1,11 @@
+## Basic documentation
+To run, simply do `docker compose up -d` in the tts and ffmpeg folders.
+This will build the container if it isn't build already, but if it is, then it'll re-use the built image.
+
+To build the container after making any changes to the non-persistent files, you can do `docker compose build`
+
+### If you are testing on local
+Once it's running, edit your config so that `TTS_HTTP_URL` is set to http://localhost:5002 and `TTS_HTTP_TOKEN` is set to `coolio`
+
+### If you are deploying to prod
+Edit your config so that `TTS_HTTP_URL` is a http request to your TTS server (whether that be localhost, an ip address or a domain) on port 5002 and `TTS_HTTP_TOKEN` is set to a random string value. You'll also need to modify the `tts-api.py` file and set the `authorization_token` variable to whatever you've set your `TTS_HTTP_TOKEN` to. This is to prevent unauthorized requests.

tools/tts/tts-api/Dockerfile

@@ -0,0 +1,40 @@
+# Grab debian-bullseye-slim
+FROM debian:bullseye-slim
+
+# install required packages
+RUN apt-get update && apt-get upgrade && apt-get install -y ffmpeg wget curl &&\
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Anaconda
+ENV CONDA_DIR /opt/conda
+RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-py310_22.11.1-1-Linux-x86_64.sh -O ~/miniconda.sh && \
+     /bin/bash ~/miniconda.sh -b -p /opt/conda
+
+# Put conda in path so we can use conda
+ENV PATH=$CONDA_DIR/bin:$PATH
+
+# Enable intel packages
+RUN conda config --add channels intel
+# Setup intel enhanced python environment named 'intel'.
+RUN conda create -n intel intelpython3_full python=3.9 numba=0.55.1 && conda clean --index-cache --tarballs --packages
+# Use the intel enhanced python environment from now on.
+SHELL ["conda", "run", "-n", "intel", "/bin/bash", "-c"]
+
+# Setup python requirements and install the TTS python module into the new intel anaconda environment.
+RUN pip install Flask &&\
+	pip install waitress &&\
+	pip cache purge
+
+COPY . /root
+RUN mkdir /tts_files
+WORKDIR /root
+
+
+# fix: `libstdc++.so.6: version `GLIBCXX_3.4.29' not found`
+# tl;dr is we need to have /opt/conda/lib in the library path for python, but *not things ran by python*
+# the python script will look for TTS_LD_LIBRARY_PATH and use it for things it runs.
+RUN conda env config vars set TTS_LD_LIBRARY_PATH=$LD_LIBRARY_PATH
+RUN conda env config vars set LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_DIR/lib
+
+ENTRYPOINT ["conda", "run", "--no-capture-output", "-n", "intel", "python", "tts-api.py"]

tools/tts/tts-api/tts-api.py

@@ -0,0 +1,54 @@
+import os
+import io
+import gc
+import subprocess
+import requests
+from flask import Flask, request, send_file, abort
+
+app = Flask(__name__)
+
+authorization_token = os.get_env("TTS_AUTHORIZATION_TOKEN", "coolio")
+
+@app.route("/tts")
+def text_to_speech():
+	if authorization_token != request.headers.get("Authorization", ""):
+		abort(401)
+
+	voice = request.args.get("voice", '')
+	text = request.json.get("text", '')
+
+	filter_complex = request.args.get("filter", '')
+	filter_complex = filter_complex.replace("\"", "")
+
+	response = requests.get(f"http://tts-container:5003/generate-tts", json={ 'text': text, 'voice': voice })
+	if response.status_code != 200:
+		abort(500)
+
+	ffmpeg_result = None
+	if filter_complex != "":
+		ffmpeg_result = subprocess.run(["ffmpeg", "-f", "wav", "-i", "pipe:0", "-filter_complex", filter_complex, "-c:a", "libvorbis", "-b:a", "64k", "-f", "ogg", "pipe:1"], input=response.content, capture_output = True)
+	else:
+		ffmpeg_result = subprocess.run(["ffmpeg", "-f", "wav", "-i", "pipe:0", "-c:a", "libvorbis", "-b:a", "64k", "-f", "ogg", "pipe:1"], input=response.content, capture_output = True)
+	print(f"ffmpeg result size: {len(ffmpeg_result.stdout)} stderr = \n{ffmpeg_result.stderr.decode()}")
+
+	return send_file(io.BytesIO(ffmpeg_result.stdout), as_attachment=True, download_name='identifier.ogg', mimetype="audio/ogg")
+
+
+@app.route("/tts-voices")
+def voices_list():
+	if authorization_token != request.headers.get("Authorization", ""):
+		abort(401)
+
+	response = requests.get(f"http://tts-container:5003/tts-voices")
+	return response.content
+
+@app.route("/health-check")
+def tts_health_check():
+	gc.collect()
+	return "OK", 200
+
+if __name__ == "__main__":
+	if os.getenv('TTS_LD_LIBRARY_PATH', "") != "":
+		os.putenv('LD_LIBRARY_PATH', os.getenv('TTS_LD_LIBRARY_PATH'))
+	from waitress import serve
+	serve(app, host="0.0.0.0", port=5002, threads=2, backlog=8, connection_limit=24, channel_timeout=10)

tools/tts/tts/.gitignore

@@ -0,0 +1 @@
+persistent_data

tools/tts/tts/Dockerfile

@@ -0,0 +1,43 @@
+# Grab debian-bullseye-slim
+FROM debian:bullseye-slim
+
+# install required packages
+RUN apt-get update && apt-get upgrade && apt-get install -y wget curl espeak-ng &&\
+    apt-get clean && \
+    rm -rf /var/lib/apt/lists/*
+
+# Install Anaconda
+ENV CONDA_DIR /opt/conda
+RUN wget --quiet https://repo.anaconda.com/miniconda/Miniconda3-py310_22.11.1-1-Linux-x86_64.sh -O ~/miniconda.sh && \
+     /bin/bash ~/miniconda.sh -b -p /opt/conda
+
+# Put conda in path so we can use conda
+ENV PATH=$CONDA_DIR/bin:$PATH
+
+# Enable intel packages
+RUN conda config --add channels intel
+# Setup intel enhanced python environment named 'intel'.
+RUN conda create -n intel intelpython3_full python=3.9 numba=0.55.1 && conda clean --index-cache --tarballs --packages
+# Use the intel enhanced python environment from now on.
+SHELL ["conda", "run", "-n", "intel", "/bin/bash", "-c"]
+
+# Setup python requirements and install the TTS python module into the new intel anaconda environment.
+RUN pip install Flask &&\
+	pip install waitress &&\
+	pip install llvmlite --ignore-installed &&\
+	pip install torch torchaudio --extra-index-url https://download.pytorch.org/whl/cu118 &&\
+	pip install TTS &&\
+	pip cache purge
+
+COPY . /root
+RUN mkdir /tts_data
+WORKDIR /root
+
+
+# fix: `libstdc++.so.6: version `GLIBCXX_3.4.29' not found`
+# tl;dr is we need to have /opt/conda/lib in the library path for python, but *not things ran by python*
+# the python script will look for TTS_LD_LIBRARY_PATH and use it for things it runs.
+RUN conda env config vars set TTS_LD_LIBRARY_PATH=$LD_LIBRARY_PATH
+RUN conda env config vars set LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$CONDA_DIR/lib
+
+ENTRYPOINT ["conda", "run", "--no-capture-output", "-n", "intel", "python", "tts.py"]

tools/tts/tts/tts.py

@@ -0,0 +1,60 @@
+import torch
+from TTS.api import TTS
+import os
+import io
+import json
+import gc
+from flask import Flask, request, send_file
+
+tts = TTS("tts_models/en/vctk/vits", progress_bar=False, gpu=False)
+
+app = Flask(__name__)
+
+voice_name_mapping = {}
+use_voice_name_mapping = True
+with open("./tts_voices_mapping.json", "r") as file:
+    voice_name_mapping = json.load(file)
+    if len(voice_name_mapping) == 0:
+        use_voice_name_mapping = False
+
+voice_name_mapping_reversed = {v: k for k, v in voice_name_mapping.items()}
+
+request_count = 0
+
+@app.route("/generate-tts")
+def text_to_speech():
+	global request_count
+	text = request.json.get("text", "")
+	voice = request.json.get("voice", "")
+	if use_voice_name_mapping:
+		voice = voice_name_mapping_reversed[voice]
+
+	result = None
+	with io.BytesIO() as data_bytes:
+		with torch.no_grad():
+			tts.tts_to_file(text=text, speaker=voice, file_path=data_bytes)
+		result = send_file(io.BytesIO(data_bytes.getvalue()), mimetype="audio/wav")
+	request_count += 1
+	return result
+
+@app.route("/tts-voices")
+def voices_list():
+	if use_voice_name_mapping:
+		data = list(voice_name_mapping.values())
+		data.sort()
+		return json.dumps(data)
+	else:
+		return json.dumps(tts.voices)
+
+@app.route("/health-check")
+def tts_health_check():
+	gc.collect()
+	if request_count > 2048:
+		return f"EXPIRED: {request_count}", 500
+	return f"OK: {request_count}", 200
+
+if __name__ == "__main__":
+	if os.getenv('TTS_LD_LIBRARY_PATH', "") != "":
+		os.putenv('LD_LIBRARY_PATH', os.getenv('TTS_LD_LIBRARY_PATH'))
+	from waitress import serve
+	serve(app, host="0.0.0.0", port=5003, threads=4, backlog=8, connection_limit=24, channel_timeout=10)

tools/tts/tts/tts_voices_mapping.json

@@ -0,0 +1,67 @@
+{
+	"p226": "Male 01",
+	"p228": "Male 02",
+	"p229": "Male 03",
+	"p230": "Male 04",
+	"p231": "Male 05",
+	"p232": "Male 06",
+	"p233": "Male 07",
+	"p234": "Male 08",
+	"p236": "Male 09",
+	"p238": "Male 10",
+	"p239": "Male 11",
+	"p241": "Male 12",
+	"p251": "Male 13",
+	"p252": "Male 14",
+	"p253": "Male 15",
+	"p254": "Male 16",
+	"p255": "Male 17",
+	"p256": "Male 18",
+	"p257": "Male 19",
+	"p258": "Male 20",
+	"p262": "Male 21",
+	"p264": "Male 22",
+	"p265": "Male 23",
+	"p266": "Male 24",
+	"p267": "Male 25",
+	"p269": "Male 26",
+	"p272": "Male 27",
+	"p279": "Male 28",
+	"p281": "Male 29",
+	"p285": "Male 30",
+	"p286": "Male 31",
+	"p287": "Male 32",
+	"p298": "Male 33",
+	"p225": "Female 01",
+	"p227": "Female 02",
+	"p237": "Female 03",
+	"p240": "Female 04",
+	"p243": "Female 05",
+	"p244": "Female 06",
+	"p245": "Female 07",
+	"p246": "Female 08",
+	"p247": "Female 09",
+	"p248": "Female 10",
+	"p249": "Female 11",
+	"p250": "Female 12",
+	"p259": "Female 13",
+	"p260": "Female 14",
+	"p261": "Female 15",
+	"p263": "Female 16",
+	"p270": "Female 17",
+	"p271": "Female 18",
+	"p273": "Female 19",
+	"p274": "Female 20",
+	"p275": "Female 21",
+	"p276": "Female 22",
+	"p277": "Female 23",
+	"p278": "Female 24",
+	"p280": "Female 25",
+	"p283": "Female 26",
+	"p284": "Female 27",
+	"p288": "Female 28",
+	"p293": "Female 29",
+	"p294": "Female 30",
+	"p295": "Female 31",
+	"p297": "Female 32"
+}