Top AI Repos — open-source AI, indexed and scored
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
Top AI Repos tracks AI repositories on GitHub and answers two different questions about each one: is it moving right now, and would you bet a product on it.
CAT is more than a CRF-based ASR toolkit: it provides a complete workflow for data-efficient end-to-end ASR, supporting CTC, CTC-CRF, RNN-T, and language-model training and inference.
| Date | Stars |
|---|---|
| 2026-07-31 | 368 |
| 2026-08-06 | 368 |
| 2026-08-14 | 369 |
| 2026-08-18 | 368 |
| 2026-09-02 | 369 |
| 2026-09-20 | 369 |
Today
— stars today
This week
— stars this week
This month
+1 stars this month
Momentum
0.0
growth rate 0.00%/day
<div align="center"><img src="./assets/logo.png" width=200></div>
# CAT: CRF-based ASR Toolkit
**CAT provides a complete workflow for CRF-based data-efficient end-to-end speech recognition.**
See [What's New](docs/whatsnew.md) for recently added functionalities and features!
- [Overview](#overview)
- [Features](#features)
- [Installation](#installation)
- [Getting started](#getting-started)
- [ASR results](#asr-results)
- [Further reading](#further-reading)
## Overview
CAT aims at combining the advantages of both the hybrid and the E2E ASR approches to achieve data-efficiency, by judiciously examining the pros and cons of modularity versus unified neural network, separate optimization versus joint optimization. CAT advocates global normalization modeling and discriminative training in the framework of [Conditional Random Field](https://en.wikipedia.org/wiki/Conditional_random_field) (CRF), currently with [Connectionist Temporal Classification](https://mediatum.ub.tum.de/doc/1292048/file.pdf) (CTC) inspired state topology.
## Features
1. CAT contains a full-fledged CUDA/C/C++ implementation of CTC-CRF loss function binding to PyTorch.
2. One-stop CTC/CTC-CRF/RNN-T/LM training & inference. See the [templates](egs/TEMPLATE).
3. Flexible configuration with JSON. Check the [guideline for configuration](docs/configure_guide.md).
4. Scalable and extensible. It is easy to be extended to train tens of thousands of speech data and add new models and tasks.
## Installation
1. Dependencies
- CUDA compatible device, NVIDIA driver installed and CUDA lib available.
- PyTorch: `>=1.9.0` is required. [Installation guide from PyTorch](https://pytorch.org/get-started/locally/#start-locally)
- [Kaldi](https://github.com/kaldi-asr/kaldi) **\[optional, but recommended\]**: used for speech data preparation and some FST-related operations. This is optional for most of the basic functions. Required only when you want to conduct [CTC-CRF](egs/TEMPLATE/exp/asr-ctc-crf) training.
Besides Kaldi, you could use `torchaudio` for feature extraction. Take a look at [data.sh](egs/aishell/local/data.sh) for how to prepare data with `torchaudio`.
2. Clone and install CAT
```bash
git clone https://github.com/thu-spmi/CAT.git && cd CAT
# Get installation helping message
./install.sh -h
# Install with default configurations
#./install.sh
```
## Getting started
To get started with this project, please refer to [TEMPLATE](egs/TEMPLATE/README.md) for tutorial.
## ASR results
| dataset | evaluation sets | performance |
| ---------------------------------------------------------------------------------------------------------------------- | ----------------------- | ------------ |
| [AISHELL-1](egs/aishell#result) | dev / test | 3.93 / 4.22 |
| [Commonvoice German](https://github.com/thu-spmi/CAT/blob/v2/egs/commonvoice/RESULT.md#conformertransformer-rescoring) | test | 9.8 |
| [Librispeech](egs/libri#result) | test-clean / test-other | 1.94 / 4.39 |
| [Switchboard](https://github.com/thu-spmi/CAT/blob/v2/egs/swbd/RESULT.md#conformertransformer-rescoring) | switchboard / callhome | 6.9 / 14.5 |
| [THCHS30](https://github.com/thu-spmi/CAT/blob/v2/egs/thchs30/RESULT.md#vgg-blstm) | test | 6.01 |
| [Wenetspeech](egs/wenetspeech#result) | test-net / test-meeting | 9.32 / 14.66 |
| [WSJ](egs/wsj/RESULT.md) | eval92 / dev93 Excerpt of 4,872 characters
Read on GitHubWould you bet a product on this? Bounded 0–100 and slow moving.
matched fp:ba24b1e48a091209, llm:Repository description: 'CAT is more than a CRF-based ASR toolkit: it provides a complete workflow for data-efficient end-to-end ASR, supporting CTC, CTC-CRF, RNN-T, and language-model training and inference.' Language: Python. No topics.
matched fp:ba24b1e48a091209, llm:Repository description: 'CAT is more than a CRF-based ASR toolkit: it provides a complete workflow for data-efficient end-to-end ASR, supporting CTC, CTC-CRF, RNN-T, and language-model training and inference.' Language: Python. No topics.
matched fp:ba24b1e48a091209, llm:Repository description: 'CAT is more than a CRF-based ASR toolkit: it provides a complete workflow for data-efficient end-to-end ASR, supporting CTC, CTC-CRF, RNN-T, and language-model training and inference.' Language: Python. No topics.