Whisper on our GPUs · diarization included · nothing to install

OpenAI Whisper, online nothing to install.

Whisper is the best open-source speech model there is — and getting it running means Python, ffmpeg, model weights, and a GPU you may not have. This is the same model, already running. Paste a link or drop a file and read the transcript in your browser.

from $0.36/hour · no subscription · Credits never expire · Never used to train AI

✓Same Whisper model · ✓No install, no Python · ✓Speaker labels included · ✓Never used to train AI

What you get

What running Whisper yourself doesn't give you.

No setup at all

The pip install is the easy part. It's the CUDA version mismatch, the missing ffmpeg, the 3 GB of model weights, and the CPU run that takes longer than the audio that stop people. None of that exists here — the model is already loaded on our GPUs.

Speaker labels Whisper can't do

Vanilla Whisper returns one undifferentiated wall of text; it has no concept of who is speaking. We run pyannote diarization alongside it, so a two-host podcast or a panel comes back as SPEAKER 1 / SPEAKER 2, already split into turns.

Finished files, not a JSON blob

The Python package hands you a dict. You get TXT, SRT, VTT, DOCX, and JSON — word-level-timed and ready to drop straight into YouTube Studio, Premiere, Resolve, or a Word document.

Your audio isn't training data

We run Whisper on Neugence-owned GPUs. Your audio and transcripts are never used to train any model, ours or anyone else's — the same privacy posture that made you consider running it locally in the first place.

Local Whisper vs Whipscribe

The same model, minus the afternoon.

✗ Running Whisper locally

Free, private, and completely yours — once it works. Budget an afternoon the first time, and a slow run every time after that if you don't have a GPU.

Python, ffmpeg, and model weights to install
CPU inference is often slower than real time
No speaker diarization at all
Output is a dict — you write the SRT formatter
Re-run the whole setup on every new machine

✓ Whisper on Whipscribe

The same Whisper weights, already warm on a GPU, with diarization and exports wrapped around them.

Nothing to install — it runs in the browser
GPU inference: most hours finish in minutes
pyannote speaker labels on every job
TXT, SRT, VTT, DOCX, JSON out of the box
Works from any machine, including a phone

Sample output

Speaker-labelled. Click-to-seek. Exportable.

Whisper's text, with the speaker turns and timings the raw model leaves out. Click any word to jump the audio.

transcript · whipscribe.com/view/whisper-online

SPEAKER 1 00:00:04 The thing people get wrong about local models is assuming the hard part is accuracy. It isn't — the weights are excellent.

SPEAKER 2 00:00:12 Right, the hard part is everything around them. Packaging, drivers, the fact that your laptop fan spins up for forty minutes.

SPEAKER 1 00:00:21 And you still don't know who said what, because the model genuinely has no idea there were two of us.

SPEAKER 2 00:00:29 That's the piece that surprises everyone the first time they run it.

Export

One transcript. Five clean formats.

Every pack exports all five — no format is held back for a higher tier.

.txt

Plain text

De-ummed paragraphs. Ready to paste.

.srt

SRT captions

Word-level. Every video editor reads this.

.vtt

WebVTT

HTML5 player + YouTube uploads.

.docx

Show notes

Formatted with chapters and pull-quotes.

.json

Machine-readable

Per-word timing + speaker IDs.

Pricing

Honest pricing, no surprises.

No subscription and no auto-renew — every tier is a one-time minute pack, and credits never expire. Buy once, use the minutes whenever you need them.

Casual pack

$4/300 min

5 hours of audio. One-time — credits never expire.

300 minutes of audio
All export formats
Unlimited Claude chat per transcript
Executive summary & key quote

Get 300 minutes

Best value

$12/2,000 min

33 hours of audio — $0.36 an hour. Credits never expire.

2,000 minutes of audio
Priority queue
All export formats
Unlimited Claude chat per transcript

Get 2,000 minutes

Bulk pack

$24/5,000 min

83 hours of audio — $0.29 an hour. Credits never expire.

5,000 minutes of audio
Priority queue
API access for batch pipelines
All export formats

Get 5,000 minutes

FAQ

Running Whisper online, answered.

Is this the real OpenAI Whisper?

Yes — the same open-source Whisper model, running on our own GPUs rather than on your machine or through OpenAI's API. Accuracy is Whisper-class: typically under 5% word error rate on clean English audio, higher on noisy field recordings or heavy accents.

Why not just run it locally?

You should, if you have a GPU and enjoy the setup — it's free and completely private. Most people trying it discover CPU inference is slower than the audio itself, that there's no speaker diarization, and that they still have to write their own subtitle formatter. This is the same model with those three problems already solved.

Which Whisper model size do you use?

We run a large-class Whisper backbone with a faster inference runtime, which is why an hour of audio typically comes back in a few minutes rather than in real time. You don't pick a model size or a runtime flag — you get the accurate one.

Do you support languages other than English?

Yes — 99+ languages, auto-detected from the audio. You don't have to tell us which language it is. Accuracy varies by language the same way it does in Whisper itself; the high-resource European languages are excellent.

What does it cost?

One-time minute packs: $4 for 300 minutes, $12 for 2,000, $24 for 5,000. Credits never expire and there's no subscription.

How large a file can I send?

Up to 5 GB and 12 hours per file through the web UI. You can also paste a YouTube, podcast, or direct media URL and we'll fetch the audio server-side — nothing to download first.

Skip the setup. Run Whisper now.

Try Whipscribe

Operated by Neugence Technology Pvt. Ltd. · contact@neugence.ai · Security · Privacy · Terms