OpenAI Whisper, without the terminal.
Speecho runs on OpenAI’s Whisper. Upload a file in your browser and get the same transcription you would get from a local install — with no pip, no CUDA, and no evening spent on dependency errors.
Transcribe with Whisper →What running Whisper yourself actually costs
The model is free. Everything around it is not.
The setup
Python, ffmpeg, a virtual environment and the right torch build for your GPU. An afternoon of dependency errors before the first transcript.
Here: open the page, drop the file. Nothing to install and nothing to keep updated.
The hardware
On a CPU, the large model runs slower than real time — an hour of audio can take well over an hour, with the fan at full tilt.
Here: it runs on server GPUs. An hour of audio comes back in minutes while you do something else.
The raw output
Whisper gives you text. Speaker labels, subtitle files, summaries and translation are each a separate project to build.
Here: all four come back from the same upload, already done.
What you get on top of the model
The parts people end up writing themselves after the first local run.
Speaker labels
Each line attributed to a voice on multi-speaker recordings — interviews, panels, meetings. Rename "Speaker 1" to real names afterwards.
Subtitles, ready to use
Export .srt and .vtt with sensible line breaks and timing, instead of post-processing Whisper’s raw segments into something a player accepts.
99 languages in, 18 out
Whisper’s multilingual strength, plus translation of the finished transcript into 18 languages and an AI summary of what was said.
Pay only for what you use
Top up once. Credits never expire. No subscription. One wallet for transcription, scans and captions.
New accounts get 15 free minutes to try — credits work for scans and captions too.
Frequently asked questions
Is this really OpenAI Whisper?
Yes. Speecho transcribes with Whisper — the same model family you would install locally, running on hosted infrastructure. You are not getting a different engine behind a Whisper-shaped label.
Should I just run Whisper locally instead?
If you have a capable GPU, you are comfortable with Python, and you transcribe constantly, then honestly yes — local Whisper costs electricity and your setup time, and nothing per hour after that. Hosted makes sense when you don’t have the hardware, don’t want to maintain the stack, or want speaker labels, subtitles, summaries and translation without building them yourself.
Do I need to install anything at all?
No. It runs in the browser — no Python, no ffmpeg, no model weights to download. Video files have their audio extracted locally before upload, so you are not sending gigabytes across the network either.
Which languages does it handle?
Whisper transcribes 99 languages, and Speecho translates the finished transcript into 18. Accuracy is strongest on clean audio in widely spoken languages, as it is with any local Whisper install.
Is my audio used to train AI models?
No. Recordings are not used for model training, source files are deleted after processing, and you can delete transcripts from your history at any time.
What does it cost?
Between $0.83 and $1.25 per hour of audio, depending on the model tier you pick. There is no subscription, credits never expire, and you get 15 free minutes when you confirm your email.
Whisper, minus the afternoon of setup.
Transcript, speaker labels, subtitles, summary and translation from one upload.
Transcribe with Whisper →