#163 · Primary category: Speech & Audio
CTCWordBeamSearch
Connectionist Temporal Classification (CTC) decoder with dictionary and language model.
Project last updated:01/31/26
GitHub Stars
579
Forks
160
Contributors
3
License
MIT
Why we included this project
Many CTC-based recognizers default to greedy decoding or a plain beam search, which can produce misspelled or out-of-vocabulary text. This decoder instead restricts output to words from a dictionary you supply and ranks candidate sequences with a language model trained on your own corpus, so the final text stays within a known vocabulary. That fits speech transcription and handwritten form reading where the valid word set is already defined. The Python package installs and passes its tests without fuss, and the README walks through a minimal example that shows how to feed a model's probability matrix and get decoded strings back. If you are building a recognizer and want to avoid post-processing guesswork, this is a straightforward component to add.
Articles for this project
No articles for this project yet.
To suggest a topic or contribute an article, contact us.
Related projects in this category
whisper.cpp
Port of OpenAI's Whisper model in C/C++
Real-Time-Voice-Cloning
Clone a voice in 5 seconds to generate arbitrary speech in real-time
VibeVoice
Open-Source Frontier Voice AI
voicebox
The open-source AI voice studio. Clone, dictate, create.
TTS
🐸💬 - a deep learning toolkit for Text-to-Speech, battle-tested in research and production