A grad student's ASR notes, frozen in time

What it does This is a personal knowledge dump from a Chinese master’s student learning automatic speech recognition. It bundles their GMM-HMM and NN-HMM Kaldi experiments, collected papers, Google’s 2018 INTERSPEECH slides, and a hand-drawn deep learning architecture cheat sheet. Think of it as finding a well-organized classmate’s binder left behind in the library.

The interesting bit The repo is explicitly unmaintained—the author graduated and moved to a WeChat group and commercial site (meta-speech.com). That makes this a time capsule of 2018–2020 ASR pedagogy, complete with the now-rare sight of someone earnestly documenting Kaldi syllable, word, and phone modeling in parallel repos.

Key highlights

• Curated Google INTERSPEECH 2018 slides, praised by the author as “very systematic”
• Personal PPTs walking through Kaldi-based GMM-HMM → NN-HMM pipeline
• A nn.png architecture diagram linking to three companion repos (syllable, word, phone modeling)
• ~618 stars suggest it struck a chord with Chinese-speaking ASR students
• Author’s CSDN blog linked for ongoing (though now also stale) writeups

Caveats

• Explicitly unmaintained: README banner says “no longer maintained” and redirects to external site
• Companion repos (ASR_Syllable, ASR_WORD, ASR_Phone) appear abandoned; one has a copy-paste error in description (“字” listed for both word and phone)
• No code in this repo itself—it’s documentation and links only
• Language is entirely Chinese; English speakers get little value

Verdict Worth a bookmark if you’re a Chinese-speaking student trying to make sense of Kaldi-era ASR and want to see how someone else muddled through. Skip it if you need runnable code, modern frameworks, or anything maintained past roughly 2020.