Our take
Activeregular commits and releases- Activity
- 75
- Community
- 94
- Issues
- 54
- Pull requests
- 47
+150 stars in the last 29 days
- Merges more pull requests than 81% of the projects we track
- More contributors than 78% of the projects we track
- Slower to merge pull requests than 86% of the projects we track
NeMo Speech is a strong, actively developed framework for teams building speech models, provided they can manage its Python and PyTorch environment and, for training, NVIDIA GPU and CUDA requirements.
Good fit if
- You need an extensible Python/PyTorch framework for ASR, TTS, or speech language model development.
- You can use pretrained checkpoints as a starting point and want to customize or deploy speech models.
- Your team can work with the GPU and CUDA setup required for training, and values an active project with frequent releases and broad contributor participation.
Look elsewhere if
- You need a turnkey hosted speech service rather than a framework for building and adapting models.
- You cannot provide an NVIDIA GPU and CUDA environment for training.
- Your project needs functionality outside the speech and audio focus of this repository.
All health signals
| Last commit | 2026-10-06 (2 days ago) |
|---|---|
| Commits, last 90 days | 100+ |
| Releases, last 12 months | 11 (latest v3.0.0, 2026-08-07) |
| Contributors | 100+ (top contributor: 11% of commits) |
| Issues closed, last 90 days | 7+ (typically closed in 6 days) |
| Pull requests merged, last 90 days | 69 (typically merged in 4 days) |
| Project age | 7 years |
Checked on 2026-10-09 with the GitHub API.
Project history
Daily GitHub snapshots since Oct 7, 2026. History covers stars and forks; health signals are not tracked historically.
Stars
18,719Forks
3,683Change since Nov 8, 2026: +5 stars·+1 forks
Overview
NVIDIA NeMo Speech is a framework for researchers and PyTorch developers building speech AI. It provides reusable model code and pretrained checkpoints for creating, customizing, and deploying systems for automatic speech recognition (ASR), text-to-speech (TTS), and speech language models.
It is a fit for teams that need a development framework rather than just a hosted speech API. The project is actively maintained, with regular releases and contributions from a broad contributor base.
Key Features
- Develop ASR, TTS, and speech language models using Python and PyTorch.
- Adapt pretrained checkpoints for new datasets or applications.
- Work with speech recognition and translation across multiple languages, as shown by the featured Parakeet and Canary models.
- Use streaming ASR models with adjustable latency options.
- Explore open-weight checkpoints and demos through the linked NVIDIA Hugging Face collection.
- Install from source with
uv, use a prebuilt NVIDIA container, or install selected toolkit extras with pip.
Use Cases
- Speech researchers can experiment with model architectures and training workflows using PyTorch.
- Product teams can adapt pretrained ASR models for transcription in their target language or domain.
- Developers building voice interfaces can evaluate streaming recognition when responsiveness matters.
- Teams creating spoken-content products can use the framework's TTS capabilities to develop speech generation workflows.
- Multilingual application teams can explore speech recognition and translation models when they need coverage beyond English.
What you need
Detected in the repository
- Python >=3.10 (from pyproject.toml)
- A test suite and automated checks on GitHub Actions
License in plain words
Apache-2.0permissive
- Commercial use: yes
- Modify and redistribute: yes
- You must keep: the license, the NOTICE file and a note of your changes
- Share your changes: no
- Includes an explicit patent grant from the contributors:
A summary, not legal advice: the LICENSE file is what applies.
Getting Started
The README recommends installing from source with uv:
git clone https://github.com/NVIDIA-NeMo/Speech.git
cd Speech
uv sync --extra all --extra cu13
See the README for installation alternatives, configuration details, and documentation links.
Considerations
- Training requires an NVIDIA GPU and CUDA; a GPU is recommended, but not stated as required, for inference. Confirm that your hardware and software stack match your intended workload.
- The README's recommended source installation uses a pinned, actively tested environment. Using a different Python, PyTorch, or CUDA stack is supported through pip, but the README distinguishes that path from reproducing the tested baseline.
- This is a fast-moving project: it has recent commits and releases, so model interfaces and setup guidance may change between versions. Pin a release or environment for reproducible work.
- The project uses Apache-2.0. Review the license and any separate terms associated with the checkpoints or services you choose to use.
- The project reports broad contributor participation and regular issue and pull-request handling; contributor concentration is not a flagged dependency risk.
Found this useful?
Share it with someone who would like Speech.