SillyTavern can read character replies aloud, but the voice has to come from a TTS provider. On a Mac, OpenVox can fill that role: it runs the speech model locally and exposes an OpenAI-compatible speech endpoint that SillyTavern can call. With Pocket TTS, you can save a cloned voice in OpenVox and assign it to a character in SillyTavern.
“Unlimited” here means OpenVox Pro removes OpenVox’s daily generation and voice clone limits on the purchased platform. The Free tier has a daily character allowance. Local generation avoids a per-request cloud speech fee, but your Mac still needs time and resources to synthesize each reply. Your SillyTavern chat model may have separate costs or limits.
The connection at a glance
SillyTavern reply → OpenVox at 127.0.0.1:8001 → Pocket TTS saved voice → spoken audio in chat
Port 8001 is the example shown in OpenVox’s Connect SillyTavern screen. Use the port your own OpenVox app displays if you changed it.
What you need for local SillyTavern TTS on Mac
- A compatible Mac with OpenVox installed and SillyTavern running on the same Mac.
- Pocket TTS downloaded in OpenVox for the cloned-voice path.
- A clean voice recording that you own or have permission to clone.
- OpenVox Pro if you want generation beyond the Free daily allowance.
You can first test the connection with a ready-made Kokoro or Supertonic 3 voice. Choose Pocket TTS when you want to use a saved cloned voice, as indicated by OpenVox’s SillyTavern setup screen.
| Model in OpenVox | When to choose it for SillyTavern |
|---|---|
| Pocket TTS | Use the saved cloned-voice workflow covered in this guide. |
| Kokoro | Test the local connection with a ready-made voice. |
| Supertonic 3 | Try another low-latency local speech option from the setup helper. |
Step 1: create and save a voice in OpenVox
- Open Voice Studio → Voice Clone in OpenVox.
- Select Pocket TTS, then provide a clear sample of one speaker. Follow the prompts in your installed app to create the voice.
- Listen to a short test sentence and save the clone with a name you will recognize in SillyTavern.
A quiet recording with no music, overlapping speech, or heavy room echo gives the model a better reference. For recording tips, see our voice cloning guide. Only clone your own voice or one you have explicit permission to use.
Step 2: start the OpenVox local TTS API
- In OpenVox, open Local API + MCP and start the server.
- Check the Base URL and Runtime Status. In the pictured setup, the base URL is
http://127.0.0.1:8001/v1. - Click SillyTavern Guide to open the Connect SillyTavern window, then select the Pocket TTS tab.
- Keep that window open so you can copy its Provider Endpoint, Model, and Available Voices values exactly.
The speech endpoint for the pictured port is http://127.0.0.1:8001/v1/audio/speech. This is the Provider Endpoint SillyTavern needs. The shorter /v1 URL is OpenVox’s API base URL. SillyTavern commonly runs on port 8000, so give the two apps distinct ports. Both can use 127.0.0.1 when they run on the same Mac.
Step 3: configure SillyTavern’s OpenAI Compatible TTS provider
In SillyTavern, open Extensions → TTS, select OpenAI Compatible, and turn on Enabled. Copy the values from OpenVox’s Pocket TTS tab into the matching fields:
| SillyTavern field | What to enter |
|---|---|
| Provider Endpoint | The full OpenVox URL ending in /v1/audio/speech |
| Model | pocket-tts in the pictured setup; copy the value from OpenVox’s Pocket TTS tab |
| Available Voices | The comma-separated list from that tab, including your saved voice |
| API Key | Leave blank for the local OpenVox server |
| TTS Speed | Start at 1.0 |
Copy the model and voice names from the app rather than guessing them; those values depend on the installed model and saved voices. After entering the fields, click Apply, then Reload or Refresh beside the TTS provider to populate the voice menus. SillyTavern’s TTS documentation notes that OpenAI-compatible providers may require you to enter the available voices manually.
Saved voices may appear as identifiers rather than the friendly names you chose. The settings screenshot shows this in the voice dropdowns. Use an identifier from your own OpenVox voice list and test it before assigning it to a character.
Step 4: map your character’s voice and test it
- Open a chat with the character you want to voice.
- In SillyTavern’s TTS voice controls or Voice Map, assign that character your saved Pocket TTS voice.
- Click Apply after editing the voice map.
- Generate a short reply and play it with the message’s speaker icon.
- Turn on Auto Generation if you want new replies to play automatically.
If you switch the model or add another cloned voice later, update Available Voices and reload the SillyTavern TTS provider again. SillyTavern’s voice-map instructions also cover assigning different voices to different characters and user personas.
Choose whether to read dialogue or the whole reply
The SillyTavern screenshot has Only narrate "quotes" enabled. With that setting, text outside quotation marks is omitted from speech. For a simple first test, use a reply such as "Hello, traveler. Welcome back." so there is quoted dialogue to read.
Turn that option off if you want the entire reply spoken. To leave out roleplay actions between asterisks, enable Ignore *text, even "quotes", inside asterisks*. SillyTavern’s narration settings explain how the two options interact. Shorter spoken replies also make it easier to judge how quickly your Mac generates audio.
Troubleshoot SillyTavern TTS connection and playback
- Connection error: Confirm that OpenVox shows the server as running and that the endpoint uses OpenVox’s port and ends in
/v1/audio/speech. - No cloned voice in the dropdown: Save the voice in OpenVox, select Pocket TTS in its setup helper, paste the current Available Voices list, and reload SillyTavern.
- No automatic playback: Check Enabled and Auto Generation, then apply the settings and voice map.
- A reply is skipped: If Only narrate "quotes" is enabled, confirm that the reply contains dialogue inside quotation marks.
- Wrong voice or pronunciation: Confirm the character’s Voice Map entry and test a cleaner reference recording.
- Using SillyTavern on another device:
127.0.0.1points to that device, not your Mac. This guide assumes both apps run on the same Mac.
Keep the voice local and use it responsibly
OpenVox generates the speech on your Mac after model setup, so the OpenVox speech step does not require a cloud TTS provider. Keep the server on loopback for this same-Mac setup. The privacy of your chat text also depends on the language model provider you choose in SillyTavern. For consent and permitted reuse, read our voice cloning ethics guide.
Frequently asked questions about SillyTavern TTS on Mac
Do I need an OpenAI API key to use OpenVox?
Leave the API key blank for the local OpenVox server. The OpenAI Compatible provider sends speech requests to your configured OpenVox endpoint. Keep your SillyTavern language-model connection configured separately.
Is unlimited SillyTavern TTS free with OpenVox?
The Free tier has a daily allowance. OpenVox Pro removes the app’s daily generation limit on the purchased platform. Speech is generated on your Mac, so speed still depends on the model, reply length, and available device resources.
Which model should I use for saved cloned voices?
Select Pocket TTS in OpenVox’s SillyTavern helper, then copy its model value and voice list. In the provided SillyTavern screenshot, the model value is pocket-tts. For broader model choices, see our local TTS model guide.
Can I keep speech generation private if my chat model is in the cloud?
OpenVox generates the audio locally. Your chat text is still handled according to the language-model provider you connect to SillyTavern. Our local TTS API guide explains the OpenVox speech backend and its loopback connection.
