Glossary
Whisper is the open-source automatic speech recognition (ASR) model OpenAI released in September 2022. It's the inflection point that made AI transcription consumer-grade — accuracy jumped from 70-85% (pre-2020 ASR) to 95-97% on clean audio, all while supporting 50+ languages with a single model. Most modern transcription tools — including ours — run Whisper or a Whisper variant under the hood.
What makes Whisper different
Three things. (1) Scale of training data: 680,000 hours of audio scraped from the web, deliberately noisy, multilingual, and including English and 50+ other languages in one model. (2) Architecture: a standard transformer encoder-decoder, but with deliberate noise injection during training so the model handles real-world recordings instead of studio-clean ones. (3) Open weights: every variant is downloadable, runnable on a GPU you own, no API needed. That's the big shift — closed-source ASR (Google, Microsoft) was the norm until Whisper.
Whisper variants
whisper-1 is OpenAI's API-served version (we use this) — fast, supports verbose JSON with segment timestamps, $0.006/minute via API. Whisper Large v3 is the open-source flagship — needs ~10GB GPU RAM, runs locally, no API costs. distil-whisper is a 6x smaller distilled version that runs on CPU. WhisperX adds forced alignment for word-level timing.
Limitations
Whisper hallucinates on long silences (it'll invent text). It struggles with very fast speech and overlapping voices. It doesn't do speaker diarization natively — that needs a separate model. And it has a 30-second context window, so longer audio is chunked (we chunk at 10-min boundaries to be safe).
In practice
When you upload an MP3 to Transcript.you, the audio is converted to a 16kHz mel-spectrogram, sent to OpenAI's whisper-1 endpoint, and the response is parsed for text and per-segment timestamps. For a 60-minute podcast, total Whisper processing time is ~90 seconds.
Related terms
Further reading
Transcribe with Whisper
Get started — freeLast updated: September 20, 2026
Frequently Asked Questions
What does Whisper mean in transcription?
An open-source ASR model released by OpenAI in 2022. Trained on 680,000 hours of multilingual audio. Now standard infrastructure for most AI transcription tools, including Transcript.you.
Why does Whisper matter when choosing a transcription workflow?
Whisper affects how you understand transcript quality, timing, compatibility, or the technology behind speech-to-text results.
Where can I apply Whisper on Transcript.you?
Use the related workflow at /ai-transcription when you want to see how this glossary concept connects to an actual transcription task.
Browse the full transcription glossary or read the complete guide to AI transcription.