Skip to content

Latest commit

 

History

1 Commit

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Speech QA — work samples

Original transcript-review examples and a small, dependency-free QA toolkit by Jakub Halamásek.

This repository was created in September 2026 as a portfolio demonstration using AI-assisted development. Every text pair and timing fixture is synthetic. It contains no client recordings, production annotations, qualification-test answers, or proprietary instructions. The examples illustrate a review approach, not independently measured professional performance.

What to review

Run it

Requires Node.js 20 or newer. There are no runtime dependencies and no installation step.

node cli.mjs
node --test

Important interpretation

WER is (substitutions + deletions + insertions) / reference tokens. Here it ignores case and punctuation, preserves diacritics and contractions, and uses Unicode NFC. WER can exceed 100%. An empty reference with a nonempty hypothesis returns null, because the ratio is undefined. Equal-cost alignments use substitution, then deletion, then insertion as a deterministic tie-break.

WER is not a score for meaning, audio intelligibility, naturalness or human reviewer quality. For example, losing a negation can be much more consequential than a formatting mismatch. Number formatting can also change token counts without changing the intended meaning.

Timing validation detects malformed timestamps, non-positive durations, out-of-range segments and same-speaker overlap. It does not establish that a word boundary matches the recording; that requires listening and inspecting the waveform. Different speakers may overlap. All durations are expressed in seconds.

About

I work with Czech audio/transcript QA and structured content review, and hold an Mgr. in teaching mathematics and physical education from Masaryk University. I also create Czech/English content and build practical tools with AI assistance.

GitHub · LinkedIn

About

Synthetic transcript review examples, Unicode-aware word error rate, timing diagnostics, and documented project case studies.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages