Term
Whisper
Whisper is OpenAI's open speech-to-text model from 2022 — a multilingual encoder-decoder transformer that turns audio into text and ships in several sizes under an MIT license.
Whisper — explained in more detail
Whisper is an automatic speech recognition (ASR) model that OpenAI released in September 2022 under an MIT license. It was trained on roughly 680,000 hours of multilingual web audio, which makes it one of the most robust open ASR models when dealing with noise, accents and technical vocabulary. Architecturally Whisper is an encoder-decoder transformer: the encoder processes a log-mel spectrogram of the audio clip, the decoder then generates the transcript token by token. The model covers 99 languages and can translate non-English speech directly into English.
Model family
Whisper comes in five sizes: tiny (~39 MB), base (~74 MB), small (~244 MB), medium (~769 MB) and large (~1.5 GB). Larger models are more accurate, especially on technical terms, code-switched passages and quieter sections — at the cost of higher RAM use and latency. As of 2026, large-v3 is the strongest variant. For local use without a cloud API, native reimplementations are common because they need far fewer resources than the Python reference.
How it differs from related terms
Whisper is not a diarization system — separating speakers requires an additional model. It also does not produce reliable word-level timestamps (forced alignment) out of the box. On very quiet or silent stretches the model tends to hallucinate, inventing text that was never spoken — upstream normalization and an explicit language hint reduce the effect noticeably.
Entdecke mehr
AI Workflows by Keyword: How We Make Recurring Routines Enforceable
A typed keyword triggers a fixed AI routine — and every single step must be committed before the next one appears. Why that's the actual trick.
GlossarReasoning Effort
Reasoning Effort is a control parameter on modern reasoning models that sets how much internal step-by-step thinking (thinking tokens) a model spends before answering — higher levels raise quality but also latency and cost.
LexikonLLM Hallucinations — Causes and Remedies
Why LLMs confidently invent falsehoods, what types of hallucinations exist, and which remedies actually help — RAG, source enforcement, verification.