Fact · Speech recognition
30seconds
The Whisper paper says its models are trained on 30-second audio chunks and cannot take longer audio as a single input.
Checked against its source on October 17, 2026
In the source's own words
“Whisper models are trained on 30-second audio chunks and cannot consume longer audio inputs at once.”
Robust Speech Recognition via Large-Scale Weak Supervision (full text), Alec Radford et al., OpenAI (arXiv), 2022, Section 3.8, Long-form Transcription.
Open the original (DOI 10.48550/arXiv.2212.04356)- What it measures
- The paper says a buffered strategy is needed to transcribe the minutes-long or hours-long audio found in real applications, which is how the open-source package handles long recordings.
- What it does not show
- The window is a property of the original model input. Speech that is cut at a window boundary can lose context, which the paper's buffering strategy has to manage.
Source
Cited on
- How to Transcribe Audio to TextGuide
- What Is Whisper AI? OpenAI's Speech Model ExplainedGuide
- Speech to Text With Translation on Phones and ComputersGuide
- Speech Recognition Models: How Neural ASR WorksGuide
- Real-Time Speech Recognition: Streaming and LatencyGuide
- On-Device vs Cloud Speech Recognition: The TradeoffsGuide
- History of Speech Recognition: From Audrey to WhisperGuide
- Can Whisper transcribe in real time?Question
- How do you transcribe a long audio file, such as one over 30 minutes?Question
- How does speech recognition know when you stop talking?Question
- Is Whisper a large language model?Question
- Is Whisper AI free?Question