kenlm

We create n-gram language models that quantify the likelihood of various sound sequences occurring in the English language.

DeepSpeech2-style ASR on LibriSpeech with CTC and KenLM beam search (HSE DLA course). WER 24.43 on test-other

macOS 英语输入法 —— 输入建议、联想输入、数字键选词，本地 KenLM 预测，不联网。A macOS English IME with word completion, next-word prediction and a floating candidate window. Fork of mgxv/SwiftType.

Next-word prediction from context + first letter: 78.6% accuracy with a LoRA-tuned Qwen2.5-3B, Mistral-7B, and KenLM blend, plus an in-browser n-gram demo.

A more accurate spelling correction for the Ukrainian language.

Rust bindings to camel-kenlm

End-to-end English speech recognition in PyTorch from scratch: CNN + BiLSTM + CTC trained on 100h LibriSpeech. 22.6% WER greedy, 12.2% with beam search + 4-gram LM under 7 GPU-hours on a single laptop GPU.

My SMT learning journey with moses, giza, and kenlm

demo of domain corpus bootstrapping using language model perplexity

Burmese n-gram language modeling on general and domain-specific corpora using KenLM and SRILM

This repo shows how to finetune the wav2vec2.0 model along with its prerequisites.

Kikuyu ASR - speech recognition for Kikuyu using MMS-300M, QLoRA, KenLM, and ONNX

Scripts to train a n-gram language models on Wikipedia articles

Basic setup to use kenlm library in cpp

Automatic Speech Recognition using Conformer with Speech Sentiment Analysis & Text Summarizer

Create and adapt n-gram and JSGF language models, e.g. for Kaldi-ASR nnet3 chain models from Zamia-Speech

A macOS IME with a floating candidate window for word completions, spell corrections, and next-word predictions powered by KenLM n-gram models.

Developed an AI tool to automatically generate captions and transcripts for YouTube videos in 67 languages and can generate summarized texts in 133 languages.

Neural Grammatical Error Correction for Romanian using Transformer