AudioUtil TTS

[crajjjj]

View File

AudioUtil TTS

---

AudioUtilTTS speaks text in game: a mod hands it a line, and a moment later you hear it. The voice is generated on your own PC by a model inside the plugin, with no internet connection, no helper program and no GPU use.

 

Requirements:

Skyrim SE or AE with SKSE64 and Address Library. Tested on 1.6.1170.
AudioUtil 0.9.26 or newer, which does the actual playback.
ConsoleUtil Extended, only for the autts test commands.

 

Installation - install audioutiltts and model

 

Usage:

If SKSE.GetPluginVersion("AudioUtilTTS") >= 0
    AudioUtilTTS.SpeakAs("Not so loud.", "whisper")
    AudioUtilTTS.Speak("The road to Whiterun lay open.", 9)
    AudioUtilTTS.SpeakActor(akSpeaker, "Not so loud.")
EndIf
;or
SendModEvent("AudioUtilTTS_Speak", "The road to Whiterun lay open.")

 

Mods using this : 

https://www.loverslab.com/files/file/41804-crajjjj-patches-hub/  Apropos2 SE

Details here: https://github.com/crajjjj/AudioUtilTTS

---

• Submitter
  crajjjj
• Submitted
  10/10/26
• Category
  Regular Mods
• Requirements
  Skyrim SE or AE with SKSE64 and Address Library. Tested on 1.6.1170.
  AudioUtil 0.9.26 or newer,
  ConsoleUtil Extended
• Regular Edition Compatible
  No
• Install Instructions

  Installation - install audioutil and model 

 

[risarei]

This works so well, will Definity use this 

[zaira]

Custom models possible? German language in my case...

Support for OpenAI api models for example?

https://lmstudio.ai/docs/developer/openai-compat

Qwen/Qwen3-TTS-12Hz-1.7B-Base for example can create scentences with Skyrim vanilla voices in a quality on the same level than Elevenlabs

 

With Openai api support people can run the engine on a separate PC while keeping the local machine vram untouched.

Edited 15 hours ago by zaira

[crajjjj]

6 hours ago, zaira said:

Custom models possible? German language in my case...

Support for OpenAI api models for example?

https://lmstudio.ai/docs/developer/openai-compat

Qwen/Qwen3-TTS-12Hz-1.7B-Base for example can create scentences with Skyrim vanilla voices in a quality on the same level than Elevenlabs

 

With Openai api support people can run the engine on a separate PC while keeping the local machine vram untouched.

German already works, just not with the default voice. The default model (Kokoro) is English only, but the plugin also loads Piper voices, and there are German ones on the same model list: thorsten (low, medium, high), thorsten_emotional, kerstin, eva_k, karlsson, pavoque and ramona.

1. Download a voice, for example vits-piper-de_DE-thorsten-medium, from https://github.com/k2-fsa/sherpa-onnx/releases/tag/tts-models
2. Unpack it to Data\SKSE\Plugins\AudioUtilTTS\models\vits-piper-de_DE-thorsten-medium
3. In AudioUtilTTS.toml, set dir = "vits-piper-de_DE-thorsten-medium" under [model] and voice = 0 under [voice]. Everything now speaks with that one voice, and you can skip the Kokoro voice download entirely.
4. Optional: give specific roles their own voice. Voices are picked through presets, and the file already has four: narrator, female, male and whisper. To point one at another model, unpack that model next to the first and replace the voice = ... line in the preset's block with a model line:

[preset.female]
model = "vits-piper-de_DE-kerstin-low"

[preset.male]
model = "vits-piper-de_DE-thorsten-medium"

OpenAI API: not supported today, the model runs inside the plugin on the CPU. One note on the link: LM Studio's OpenAI-compatible server has no speech endpoint (only chat, completions, embeddings and so on), so LM Studio alone could not serve Qwen3-TTS to the plugin. What it would need is a server with POST /v1/audio/speech, and those exist for Qwen3-TTS (Qwen3-TTS-Openai-Fastapi, vLLM-Omni).

Maybe will  add in the future - currently focused on the dialogue system 

[fork1333]

does the mod support player character vc too? if yes what mod can cause the issue that player vc cant be played?