- Home
- Skills
- Productivity
- Transcriber by shash.builds
Transcriber by shash.builds
Upload a voice note, podcast, call or video and get a transcript and timestamped captions, made with OpenAI's Whisper inside the chat.
Free
Works with the AI Agents you already use
Transcriber by shash.builds
Example session with this skill installed
Transcribe this voice note. It's Hinglish. [uploads note.m4a]
- Read your context and instructions
- Compiled the transcriber by shash
The full transcript, Hindi in Roman letters with English words kept and doubtful words marked; the language detected and the duration; files note.txt, note.srt, note.vtt and note.json (word timings); and next steps such as captions for a reel, a post or YouTube chapters.
Connects securely to your tools. The creator never sees your data.
About this skill
Upload a voice note, podcast, call or video and get a transcript and timestamped captions, made with OpenAI's Whisper inside the chat.
English, Hindi, Hinglish and 90+ languages. The audio stays in the sandbox.
What you get
- The full transcript (a summary first for long recordings), with doubtful words marked instead of guessed
- The language detected and the duration
.txt,.srtand.vttcaptions, plus a.jsonwith word timings- Ideas for what to do next: captions for a reel, a post in your voice, YouTube chapters
Before and after
| Claude without the skill | Claude with the skill |
|---|---|
| Says it cannot listen to audio | Runs Whisper in its own sandbox and returns the text |
| Hindi comes out in Devanagari, English words garbled | A Hinglish option writes Hindi in Roman letters and keeps English words in English |
| Guesses names and numbers | Marks doubtful words and lets you pass brand names as hints |
| Gives text only | Writes .txt, .srt, .vtt and a .json with word timings |
Examples
- "Transcribe my call recording and pull out the action items." The transcript, then the list. Claude checks you have consent before you share a recording of other people.
- "Make captions for this reel." An .srt file you can upload or burn into the video.
- "Translate this Tamil voice note to English." An English translation of the speech.
- "Transcribe this with the most accuracy you can. It has prices in it." The larger model, which handles numbers and names better, with a bigger first download.
Models
base (default) and small; medium and large-v3 are more accurate and much larger downloads. The first run installs the engine and model by itself (a minute or two); later runs reuse them.
Free and MIT licensed. Part of the free shash.builds skill collection by Shashwat Jain.
Whisper mishears names, brands, numbers and mixed languages; doubtful words are marked rather than guessed. The default base model suits clear English drafts; use --size small or larger for Hindi, numbers, noisy audio or anything you will quote. The verified shash.builds model packages are not published yet, so for now the model is downloaded by faster-whisper from Hugging Face, which the sandbox must be able to reach. medium and large-v3 are about 1.5 GB and 3 GB. If packages cannot be installed, it lists phone options instead (Voice Memos, Android Recorder, YouTube auto-captions).
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Free forever
No account required to browse
Trust & safety
Security scan
Security scan passed 13 days ago
- Free with an Agensi account
Requires
Permissions
Allowed Hosts
File Scopes
Installs faster-whisper from PyPI and downloads a Whisper model on first use. The audio stays in the sandbox and is not sent to any other service.