NCKChaitanya KumarTeach · Build · Inspire
Menu
HomeBioPublicationsCoursesApplicationsResearchResources
Sign in ↗
Applications

The practical toolkit · Python 3.11+

MyToolsSmall tools.
Clearer work.

Five focused utilities to review recordings, reconcile records and keep a useful trail of evidence. Download the source, follow an example and make it part of your own workflow.

Download all five Installation guide

Source code · Examples · Documentation · Offline tests

05Independent tool folders
04Fully offline · standard library only
01Optional hosted AI integration

Start here

Download. Extract. Run locally.

This page provides the software and instructions. The tools run in a terminal on your own computer; files are not processed by this website.

  1. 1

    Install Python 3.11+

    Use Python for your operating system . Check with python --version on Windows or python3 --version on macOS/Linux.

  2. 2

    Choose a download

    Extract an individual tool ZIP or the full collection. Inside the extracted MyTools folder, open a terminal in the numbered tool folder containing tool.py.

  3. 3

    Try the included example

    Tools 01–04 need no extra packages. Tool 05 has a separate setup below. Use tool.py --help with your Python command for available options.

A few useful conventions

Commands below start inside the individual tool folder. Windows examples use python; use py -3 if that is how Python is installed. Choose a new output name for repeat runs and keep your source inputs unchanged. Each download includes tests and a complete README.

Tools 01–04 Offline · no external AI service
01Listen & review

Listening Dashboard

Bring recordings, transcripts and review notes together in one local browser page. Listen through a batch with a clear view of what is ready, under review or needs another take.

Use it forPodcast edits, voice-over takes, training recordings, interviews and accessibility narration.

Download source Full README
Input

A JSON queue + local audio or video

Output

An offline HTML listening page

How to use it

  1. 1

    Put your media in a local folder and describe each item in the example queue format.

  2. 2

    Generate the dashboard, then open the HTML file in a browser and listen.

  3. 3

    Update notes and status in the JSON queue. Rebuild the page to show your changes.

Installation & first run

Python 3.11+ and a browser are enough. From the extracted tool folder, run this included example. It creates a two-second synthetic tone; start playback at low volume.

python examples/create_demo_tone.py
python tool.py --queue examples/queue.json --media-root output --output output/listening.html

What to expectOpen output/listening.html. The example demonstrates the player and review layout. For your own files, replace the queue and media-root paths.

This is a human listening aid. It does not transcribe or assess sound. Media stays local and is linked rather than embedded; keep the HTML and media together when moving them. Use --force only when you intend to rebuild an existing output.

02Match your records

Catalog Reconciler

Connect a saved catalog to local records using stable IDs and title candidates. Keep a consistent review list as snapshots change, with original metadata and earlier decisions preserved.

Use it forDocument libraries, product catalogs, learning resources and media collections.

Download source Full README
Input

Catalog snapshot + optional local records and ID mappings

Output

A reconciled JSON catalog

How to use it

  1. 1

    Prepare a snapshot with catalog_id and items. Add local records and explicit mappings if available.

  2. 2

    Run the reconciler and inspect mapped, candidate-only and unmapped entries.

  3. 3

    Pass the prior result with --previous on the next run to preserve scope and decisions.

Installation & first run

Python 3.11+ is the only requirement. No packages or account are needed. Open a terminal in the extracted tool folder and run the neutral example.

python tool.py --snapshot examples/snapshot.json --local examples/local.json --mapping examples/mapping.json --output outputs/catalog.json

What to expectOpen outputs/catalog.json. The example returns three items: one explicitly mapped, one with a title candidate and one unmapped. Later runs can use --previous outputs/catalog.json with a new output filename.

Title matches suggest candidates; they do not prove content identity. A previous catalog freezes the scope: missing items are retained and new IDs are listed separately. Inputs remain unchanged; existing outputs require --overwrite.

03Check what changed

Metadata Verifier

Check whether observed titles and descriptions match your intended updates. Get an explicit result for every field, including the expected value and the value found in the snapshot.

Use it forContent updates, product listings, document repositories and learning-library maintenance.

Download source Full README
Input

A scoped catalog + observed snapshot + intended updates

Output

A field-by-field JSON verification report

How to use it

  1. 1

    Define the allowed item IDs in a catalog and the intended title/description changes in a batch.

  2. 2

    Export or prepare a fresh snapshot of the records you want to check.

  3. 3

    Run the comparison and inspect each mismatch or missing item in the report.

Installation & first run

Python 3.11+ is sufficient; there is no package installation. From the extracted tool folder, try the included passing example.

python tool.py --catalog examples/catalog.json --snapshot examples/snapshot.json --batch examples/batch.json --output outputs/metadata-report.json

What to expectOpen outputs/metadata-report.json. Exit code 0 means every requested field matched; 1 means mismatches were reported; 2 indicates an input or file error.

Comparison is exact after normalizing line endings. Case, punctuation and spaces still matter. Verification applies to your supplied snapshot and requested fields; it does not certify content quality or snapshot freshness. Existing reports require --overwrite.

04Keep the reasoning

Decision Tracker

Record a decision, the reason behind it and the evidence that supports it. Maintain a readable history of what to retain, hold, revisit or exclude from a review.

Use it forAsset audits, editorial queues, document-retention reviews and catalog cleanup.

Download source Full README
Input

A scoped catalog + decision batch + local evidence files

Output

An updated catalog with decision history and evidence hashes

How to use it

  1. 1

    Give each decision a status, reason, revisit condition and relative evidence-file paths.

  2. 2

    Choose the evidence directory and run the tracker against your existing catalog scope.

  3. 3

    Inspect the updated catalog. Use a new batch ID and output filename for the next review.

Installation & first run

Python 3.11+ is the only dependency. The download includes fictional notes and a batch with two decisions. Run this from the extracted tool folder.

python tool.py --catalog examples/catalog.json --batch examples/batch.json --evidence-root examples/evidence --output outputs/decisions-catalog.json

What to expectOpen outputs/decisions-catalog.json. One example item is retained and another is held. Each evidence record includes its relative path, SHA-256 hash and byte size.

Evidence must stay inside the chosen directory. A hold needs explicit unresolved checks. The tool records your judgment; it does not validate the truth of evidence or delete excluded items. Inputs and evidence stay unchanged; existing outputs require --overwrite.

05Check spoken audio
Online AI integration

Qwen Audio Checker

Collect audio-quality evidence from a public Qwen demo. Check intelligibility, pronunciation, pacing, recording quality and the ending of a spoken clip.

Use it forNarration, spoken presentations, training clips, voiceovers and podcast excerpts up to 180 seconds.

Download source Full README
Report formats Validation notes
Free demo access

The public demo is currently free to access with no user API key or paid account, as checked on 14 October 2026. Capacity, continued free access and availability are not guaranteed. Direct model APIs may be paid.

Official Qwen demo Direct API pricing
Input

A speech recording + a human-reviewed calibration clip

Output

JSON findings, raw responses and full audio-window coverage

How it works

  1. 1

    Prepare locally. Create a hashed audio bundle. Compatible PCM16 WAV needs only Python; other audio/video formats also need FFmpeg.

  2. 2

    Calibrate deliberately. Use a clear recording you have listened to. The demo assesses clean, silent and reversed controls; inspect the returned evidence.

  3. 3

    Review and listen. Submit the full clip and overlapping 12-second windows. Inspect the report and listen to uncertain passages yourself.

Install & prepare locally

Open a terminal in 05-qwen-audio-checker. Install the packages in a virtual environment. Replace recordings/narration.wav with your own speech recording, no longer than three minutes. The preparation command does not upload it.

python -m venv .venv
.\.venv\Scripts\python.exe -m pip install -r requirements.txt
.\.venv\Scripts\python.exe tool.py --help
.\.venv\Scripts\python.exe tool.py prepare recordings/narration.wav --output runs/narration-01

Online operations require httpx and gradio_client, pinned in requirements.txt. Install FFmpeg for MP3, M4A, MP4 or other inputs needing conversion. The README includes platform-specific installation steps.

What to expectruns/narration-01 contains bundle.json and audio.wav. Keep this bundle and the original file unchanged, and always choose a new output directory.

Calibrate & review online

These commands send audio outside your computer. The explicit --send-audio flag is required for calibration and review. Use recordings you are permitted to share with the official Hugging Face Space and its Alibaba backend. Review also sends any optional reference text you supplied.

Replace recordings/known-good.wav with a clip you have listened to and confirmed is clear. Aim for at least 12 seconds of uninterrupted speech and more than eight words.

.\.venv\Scripts\python.exe tool.py prepare recordings/known-good.wav --output runs/known-good
.\.venv\Scripts\python.exe tool.py calibrate runs/known-good/audio.wav --output runs/calibration-01 --known-good --send-audio

Open runs/calibration-01/report.json and inspect the recognized words and control responses. Once the calibration is suitable, review your prepared recording:

.\.venv\Scripts\python.exe tool.py review runs/narration-01/bundle.json --calibration runs/calibration-01/report.json --output runs/review-01 --send-audio

Read the resultOpen runs/review-01/report.json. Candidate evidence still needs inspection; held or failed results require attention. Every report keeps publication_approved: false.

Qwen provides the audio understanding through a third-party service. This is a review aid, not a guarantee of quality. It does not verify factual accuracy, ownership or subtitles, and it never publishes media. The pinned demo can stop working if its revision or API changes; consult the README before changing any checks.

Keep the complete toolkit

Source, examples and guides.
In one download.

Each tool works independently. Start with the task you need today and keep the rest for later.

Download MyTools