Kaldi Reader - SCP & ARK Viewer
English | 简体中文 | 日本語
Kaldi Reader lets you open Kaldi .scp files in VS Code and click .ark:<offset> references to inspect the actual entry. It is built for browsing training data, feature matrices, alignment vectors, and audio segments packed in wav ark files.
Supported Kaldi ark entries:
wav.ark:<offset>: validates RIFF/WAVE at the offset and opens the audio with AudioLens.
FloatMatrix(FM): opens the full matrix as raw text.
CompressedMatrix(CM / CM2 / CM3): decompresses the entry and opens the full matrix as raw text.
Int32Vector: opens the full integer sequence as raw text, commonly used for alignments.
Install
Install from the Visual Studio Marketplace:
https://marketplace.visualstudio.com/items?itemName=simzhou.kaldi-reader
Or from Open VSX:
https://open-vsx.org/extension/simzhou/kaldi-reader
For ordinary audio file paths such as .wav, .flac, .mp3, .pcm, and .raw, install AudioLens, the companion audio viewer extension:
https://marketplace.visualstudio.com/items?itemName=simzhou.audiolens
Usage
Open a Kaldi .scp file, for example:
utt001 /data/train/feats.ark:12345
utt002 feats.ark:67890
utt003 ali.ark:345
Kaldi Reader turns *.ark:<offset> references into clickable links. When you click a link:
- wav ark entries open in AudioLens.
- feature matrix entries open as raw matrix text.
- alignment / int-vector entries open as raw integer sequence text.
Kaldi Reader is a workspace extension and works in both local and Remote SSH workspaces. If the .scp file and the referenced .ark file are on a remote machine, parsing and reading happen in the remote extension host.
Path Resolution
Relative ark paths are resolved with two stable bases:
- The directory of the current
.scp file.
- The current workspace root.
Other implicit CWD-based paths are not guessed automatically. This avoids opening the wrong ark file in large training directories where duplicate file names are common.
Ordinary Audio Paths
Kaldi Reader only handles *.ark:<offset>. For ordinary audio paths such as:
/data/audio/utt001.wav
utt002 /data/audio/utt002.flac
Use AudioLens. AudioLens can turn ordinary audio paths in text files into clickable links.
Install From VSIX
Install a local packaged build with:
code --install-extension dist/kaldi-reader-0.1.2.vsix
Development
npm install
npm run build
npm run typecheck
npm run package
Press F5 in VS Code and choose the extension development host. Open a Kaldi .scp file and click an ark entry link.
Author
SimZhou: https://simzhou.com/en/about/
Copyright
Copyright (c) 2026 SimZhou. All rights reserved.