GLaDOS_TTS

GLaDOS_TTS

A text to speech voice named GLaDOS for NVIDIA NeMo, with the steps to convert it to Riva files and serve it on a Jetson. For people who build voice projects on NVIDIA hardware.

Made by DavesArmoury on Hugging Face. These notes are Mindket's, written from its model page and model card; the model's own files and license are on Hugging Face.

About

What it is: GLaDOS_TTS is a text to speech voice for NVIDIA NeMo, made by DavesArmoury and named GLaDOS. It reads English text aloud in that voice. The card points to the notebook used for training and walks through turning the models into files that NVIDIA Riva can serve.

Who it suits: Developers and hobbyists who run their own speech synthesis on NVIDIA hardware and want a ready-made voice named GLaDOS.

Industry | Who uses it | How they use it | Example
Hobby electronics | Jetson tinkerers | Give a project a spoken voice on a Jetson | A desk assistant that speaks its replies
Software | Voice app developers | Serve a custom voice through Riva | A prototype that reads text aloud in the GLaDOS voice
Research | Speech researchers | Study a NeMo-trained voice and its Riva conversion | Retrace the training notebook and the conversion steps

Anyone who wants a click-and-run voice app can skip it, since the card is a build recipe and not an app.

Where to use it: The voice is served through NVIDIA Riva, and the card builds it on a Jetson device.

• NVIDIA Riva speech services
• A Jetson device running Docker, where the card builds and serves the voice
• Your own computer with Python, for downloading the files and converting the models
• NeMo, the framework the voice was trained in

Use case | Who | What they get back
Speech output | Jetson projects | Spoken English replies in the GLaDOS voice
Voice serving | Developers | A Riva server that holds the GLaDOS voice
Model conversion | Speech engineers | Riva files made from the NeMo ones

How to set it up: The setup has two parts, converting on your computer and building on the Jetson.

• Python with pip
• A Jetson device with Docker and NVIDIA GPU support
• The ngc command line tool, used to fetch the Riva quick start

1. Download the model files with the hf command.
2. On your own computer, convert the hifigan and fastpitch NeMo files into Riva files with nemo2riva.
3. Copy the two .riva files to the Jetson, into its RIVA artifacts folder.
4. On the Jetson, start the servicemaker container, build the speech synthesis package and deploy it.
5. Leave the container, fetch the Riva quick start and run its init script.
6. Copy the glados .riva files into the Riva models folder.

pip install -U huggingface_hub
hf download DavesArmoury/GLaDOS_TTS

The conversion on your own computer uses these commands.

pip3 install whl
pip3 install nemo2riva

nemo2riva --out hifigan.riva hifigan.nemo --key None
nemo2riva --out fastpitch.riva fastpitch.nemo --key None

On the Jetson, the container and the build use these.

docker run --gpus all -it --rm \
    -v /home/davesarmoury/RIVA/artifacts:/servicemaker-dev \
    -v /home/davesarmoury/RIVA/riva_repo:/data \
    --entrypoint="/bin/bash" \
     nvcr.io/nvidia/riva/riva-speech:2.13.1-servicemaker-l4t-aarch64

riva-build speech_synthesis \
    /servicemaker-dev/glados.rmir:tlt_encode \
    /servicemaker-dev/glados_fastpitch.riva:tlt_encode \
    /servicemaker-dev/glados_hifigan.riva:tlt_encode \
    --voice_name=GLaDOS \
    --sample_rate 22050

riva-deploy /servicemaker-dev/glados.rmir:tlt_encode /data/models

Back outside the container, the quick start comes next.

ngc registry resource download-version nvidia/riva/riva_quickstart_arm64:2.13.1
cd riva_quickstart_arm64_v2.13.1
bash riva_init.sh

How to use it: Once the voice files sit in the Riva models folder, you start Riva with its start script. The card shows no command for sending text to the voice, so use the Riva client you already have.

riva_start.sh

Tool | What it does
nemo2riva | Converts a NeMo model file into a Riva file
riva-build | Builds the GLaDOS speech synthesis package at a 22050 sample rate from the two Riva files
riva-deploy | Deploys the built package into the models folder
riva_init.sh | Runs the Riva quick start setup
riva_start.sh | Starts Riva so it can serve the voice

Example: You ask Riva to speak a short English sentence with the voice name GLaDOS. The card gives no sample output, so what comes back is spoken audio in that voice, with quality you judge by ear.

For newcomers and experts: This is a voice model, a file set that lets a computer read text aloud in one particular voice. It is not an app, so the first small thing to try is downloading the files and seeing which NeMo files they contain.

The card lists two NeMo files, one named fastpitch and one named hifigan. It converts both to Riva format and joins them into one speech synthesis build with a 22050 sample rate. The conversion passes the key value None.

Download the NeMo files
Convert them with nemo2riva
Build with riva-build
Deploy with riva-deploy
Start Riva with riva_start.sh

Good to know: The card is short and covers the build path only.

• Language: English.
• Hardware: the card builds on a Jetson device with NVIDIA GPU support.
• Voices: if you adapt or retrain the voice, use only voices you have the right to use.
• License: CC BY 4.0.
• Cost to run: none named.
• Data it reaches: the files you download and convert. The card names no web or account access beyond fetching the Riva quick start with ngc.

On Hugging Face, 13 November 2026

• Task: text to speech
• Library: nemo
• Downloads in the last 30 days: 325
• Likes: 17
• License: cc-by-4.0
• Languages: en
• Tags: nemo, onnx, text-to-speech

Links

• Model: https://huggingface.co/DavesArmoury/GLaDOS_TTS
• Files: https://huggingface.co/DavesArmoury/GLaDOS_TTS/tree/main

Get the model

Download its files with Hugging Face's command line:

pip install -U huggingface_hub
hf download DavesArmoury/GLaDOS_TTS

Then follow the usage in its model card. Nothing in this kit runs by itself.