Parakeet Ultra Transcriptor

Parakeet Ultra Transcriptor

Turns an uploaded audio or video file into full text, with a TXT download and an SRT subtitle file when word timings allow. The interface is in Spanish, and it suits students and anyone transcribing recordings.

Made by Zangurak1 on Hugging Face. These notes are Mindket's, written from its Space page and README; the app's own files and license are on Hugging Face.

About

What it is: This Space transcribes audio or video into text. You upload a file, read the full text and download it as UTF-8 TXT, plus an SRT subtitle file when the timings are usable. It uses the moondream parakeet-ultra model with a Spanish interface and runs on zero-a10g hardware at Hugging Face.

Who it suits: It suits students and other people who need the text of a recorded class, talk or video, and developers who want to see how a transcription app handles limits and privacy.

Industry | Who uses it | How they use it | Example
Education | Students and teachers | Turn a recorded class into text for study | Upload a lecture video and download the TXT
Media | Video editors | Get subtitles from a recording | Download the SRT file for a clip
Software | Developers | Call the transcription API | Send a file to the public transcribe endpoint

You can skip it if your recordings hold sensitive content, since the page warns against uploading it.

Where to use it: You can use it in these places:

• A browser: open the Space's page on Hugging Face and nothing needs installing.
• Windows, macOS or Linux: run your own copy with Python 3.12 and ffmpeg.
• Gradio's public API: the transcribe endpoint takes one input file.

Use case | Who | What they get back
Class transcripts | Students | Full text and a TXT download
Subtitles | Editors | An SRT file built from real word timings
API transcription | Developers | Text from a file sent to the endpoint

How to set it up: You need these first for a local copy:

• Python 3.12 and pip
• ffmpeg with ffprobe on your PATH
• The system packages listed in packages.txt, which Hugging Face installs with apt-get

The online page needs none of this. For your own copy, follow the steps below.

1. Install the Hugging Face tools.
2. Download the Space's files into a folder.
3. Move into that folder.
4. Install Gradio at the version the Space uses, then the other packages.
5. Start the app.

pip install -U huggingface_hub
hf download Zangurak1/parakeet-ultra-transcriptor --repo-type space --local-dir parakeet-ultra-transcriptor
cd parakeet-ultra-transcriptor
pip install gradio==6.31.0
pip install -r requirements.txt
python app.py

The README also describes a local check on the CPU, for a copy that already has the model in its cache. It does not apply inside a Space.

PARAKEET_LOCAL_CPU=1 python app.py

How to use it: Open the Space in a browser, or start your own copy with the command above. Upload an audio or video file, wait for the text, then download the TXT or the SRT. Files can be up to 3 hours and 2 GB, but long ones may not finish in the GPU time, so split them.

Tool | What it does
Text output | Shows the full transcript on the page
TXT download | Keeps exactly the text the model returned
SRT download | Builds subtitles from real word timings, grouped by pauses and punctuation
transcribe endpoint | Gradio's public API, with one input file

Example: You upload a recording of a class. The Space converts it to mono 16 kHz audio, transcribes it and shows the full text. You then download the TXT, and the SRT too if the timings are valid.

For newcomers and experts: If you are new to this, start with a short recording of a minute or two. Read the text, download the TXT and check the result before you try a long file.

If you know the field, the audio goes through ffprobe and ffmpeg on the CPU, with no shell and only file and pipe protocols. Only inference asks for ZeroGPU, with a reservation of at most 120 seconds based on the audio length. SRT uses only the model's word timings and never invents or stretches them.

You upload audio or video
ffmpeg converts it to mono WAV
The model transcribes it on a GPU
Text and TXT are saved
SRT is built if timings are valid

Good to know: The authors tested one public 15-second example, which took 16.7 seconds in total, including connection and queue time. Long recordings are untested and may use up the reservation or your quota. The Space does not summarize your classes or send them to ChatGPT.

• License: apache-2
• Cost to run: ZeroGPU quota for each run, and no paid service is named
• Data it reaches: the audio or video you upload, kept briefly in the Gradio cache for about an hour, and Hugging Face may keep its own logs
• Not for: sensitive content, files over 3 hours or 2 GB, or private storage, since download links of a public Space are not private

On Hugging Face, 13 November 2026

• Likes: 3
• Built with: gradio
• License: apache-2.0

Links

• Space: https://huggingface.co/spaces/Zangurak1/parakeet-ultra-transcriptor
• Files: https://huggingface.co/spaces/Zangurak1/parakeet-ultra-transcriptor/tree/main

Try it

Open the Space in your browser, or copy it to your own Hugging Face account from its page. Nothing in this kit runs by itself.