# TTS

SillyTavern has a wide range of TTS options. This page explains the setup and use.

# What is it?

TTS is used to have a voice narrate parts of your chat.

# Configuring TTS

# TTS Provider Selectbox

Used to select which TTS service you want to use.

  • ElevenLabs - paid subscription required, highest quality voices available at present.
  • Silero - free, runs on your PC, quality can vary widely
  • System - uses your OS TTS engine, if one exists. Quality can vary widely depending on the OS.
  • Edge - free, runs via Azure, generally quite fast, and voices feel natural but dry and emotionless. Like listening to the evening news or a radio announcer.
  • Coqui-TTS - free, No API Implementation at this time. High-performance Text2Speech models (Tacotron, Tacotron2, Glow-TTS, SpeedySpeech) as well as Bark.
  • Novel - requires a paid NovelAI subscription, generated by NovelAI's TTS engine
  • RVC - free, voice cloning

# Checkboxes

  • Enabled - turns TTS playback on/off
  • Auto Generation - lets TTS start playing automatically when a new message enters the chat
  • Only narrate "quotes" - Limits TTS playback to only include text within "quotation marks". This will *include "quotes" within asterisk lines* (internal variable name = narrate_quoted_only)
  • Ignore *text, even "quotes", inside asterisks* - TTS will not play any text within *asterisks*, even "quotes" (internal variable name = narrate_dialogues_only)
  • having both "only narrate quotes" and "ignore asterisks" checkboxes both checked will result in the TTS only reading "quotes" which are not in asterisks, and ignoring everything else.
  • Narrate only the translated text - this will make the TTS only narrate the translated text.

Given the example text: *Cohee approaches you with a faint "nya"* "Good evening, senpai", she says. Here's a table showing how the text will be modified based on the boolean states of Ignore *text, even "quotes", inside asterisks* and Only narrate "quotes":

Ignore *text, even "quotes", inside asterisks* Only narrate "quotes" Output
Disabled Disabled Cohee approaches you with a faint "nya" "Good evening, senpai", she says.
Disabled Enabled "nya"... "Good evening, senpai"
Enabled Disabled "Good evening, senpai", she says.
Enabled Enabled "Good evening, senpai"

# Sliders

These will change depending on the API you select.

(explanation coming soon)

# Buttons

  • Apply - this must be clicked after setting a TTS API and after editing the voice map.
  • Available voices - loads a popup with all voices available for your selected API, and lets you preview them with sample dialogues.

# Using TTS

  1. Click the "Enable" checkbox, or nothing will ever happen.
  2. Click the "Auto-generation" checkbox if you want the TTS to start automatically every time a new message arrives in chat.
  3. Optionally, click the megaphone icon inside the top-right of any message to playback on demand.
  4. Click the lower right "Stop" button (found inside the wand menu) to stop any playback.

# Voice Map

You must provide a voice map for the TTS to use, otherwise, it won't know what voices should be used for each character.

These must be in the exact format stated below:

CharacterName:TTSVoice,CharacterName2:TTSVoice2

For Coqui-TTS the format needs to include the speaker and language from the WebGUI:

CharacterName:TTSVoice[speakerid][langid] or Aqua:tts_models--multilingual--multi-dataset--your_tts\model_file.pth[2][1]

# Bark ZeroShot Voice Cloning Speakers

If using Bark you must create a voice folder with a voice file to clone. Ensure you add voices to homedir\tts\bark_v0\speakers. On Windows it is probably C:\Users\USERACCOUNT\AppData\Local\tts\bark_v0\speakers\ type %appdata% in windows explorer then go UP a directory to local and you should see tts.

The directory should look like this:

  • homedir
    • tts
      • bark_v0
        • speakers
          • customvoice1
            • speaker.wav
            • speaker.npz
          • robinwilliams
            • speaker.mp3
          • me
            • speaker.mp3

One first load of this model and voice bark will clone the voice and create a .npz file, this is needed for faster TTS.