Practical overview
When it fits
Use it when an application running on a supported Jetson device needs one or more of its documented ASR, NLP or TTS capabilities.
Who it is for
- Developers building voice-enabled applications for supported NVIDIA Jetson hardware.
- Teams needing local speech recognition, NLP or text-to-speech inference on Jetson.
Documented uses
- Transcribe live audio in streaming chunks.
- Recognize commands or keywords and detect voice activity.
- Classify intent and slots, sentiment or named entities, and run question answering.
- Generate speech from text.
Requirements & limits
The project specifies Python and JetPack 4.4.1 or newer, and names Jetson Nano, TX1/TX2, Xavier NX and AGX Xavier. It is distributed as a Docker container.
- Inference uses the Jetson device's onboard GPU.
- Models are trained with NeMo and deployed with TensorRT; NLP models use DistilBERT.
- Streaming ASR transcribes in one-second chunks and uses pauses to determine sentence endings.
- TensorRT optimization can take a few minutes the first time each model runs; the optimized model is cached.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Start the Jetson Voice container
Clone the repository and launch its documented container environment.
Before you start
- A supported Jetson device: Nano, TX1/TX2, Xavier NX or AGX Xavier.
- Python and JetPack 4.4.1 or newer.
- Docker.
Steps
Clone the repository using its documented dev branch command.
Change into the jetson-voice directory.
Run docker/run.sh to start the container.
Expected outcome
The container starts in an interactive shell unless a run command is specified.
Guide limitations
- The guide covers container startup, not running a particular inference task.
- USB microphones or speakers should be connected before starting the container.
Try streaming ASR with an audio file
Run the documented ASR example on a supported audio file from inside the container.
Before you start
- The Jetson Voice container started as described in the repository README.
- A WAV, OGG or FLAC audio file available inside the container.
Steps
From the container, run examples/asr.py with the --wav option and the path to the audio file.
Read the incremental transcription printed by the example.
Expected outcome
The example prints streaming transcription in successive partial results.
Guide limitations
- TensorRT may take a few minutes to optimize a model on its first run; the optimized model is cached for later runs.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 15 October 2026
- Original website title
- dusty-nv/jetson-voice · Official homepage title · retained in its source language