voice-ai

Speech: Build AI Voices

Name: Speech: Build AI Voices
Author: NVIDIA-NeMo

Get scalable speech AI with Speech, a framework for researchers and developers, backed by 18k+ GitHub stars.

17,787 stars3,507 forksPythonHealth Score 8/10Updated 7/17/2026100% free · open source

What it does

Speech is a scalable generative AI framework for building Automatic Speech Recognition and Text-to-Speech systems using Large Language Models and Multimodal techniques.

Install / run

git clone https://github.com/NVIDIA-Nemo/Speech.git

When to use it

•When building voice assistants or voice-controlled applications
•When developing speech-to-text or text-to-speech systems for content creation or accessibility
•When researching or experimenting with multimodal AI models that combine speech, text, and other modalities

Quick start

1Navigate to the cloned repository with `cd Speech`
2Install required packages with `pip install -r requirements.txt`
3Explore the `examples` directory for sample scripts, such as `asr.py` or `tts.py`, to get started with speech recognition or synthesis
4Modify the configuration files, like `config.yaml`, to customize model settings or experiment with different architectures
5Run a sample script, for example, `python examples/asr.py --model=QuartzNet` to perform speech recognition with a pre-trained model

Ready-to-paste prompt

python examples/tts.py --model=HiFiGAN --text='Hello, how are you?' --output_file=speech_output.wav

Heads up: Requires a compatible NVIDIA GPU and the CUDA toolkit installed to utilize GPU acceleration, which is a prerequisite for many of the framework's features

Saves to your device