Whisper Base.en

Whisper base.en is an automatic speech recognition (ASR) model that transcribes English speech into text. It is a Transformer-based encoder-decoder (sequence-to-sequence) model trained on 680,000 hours of weakly supervised audio data, enabling strong generalization to many datasets and domains without fine-tuning. The model is optimized for English-only speech recognition. According to OpenAI, the English-only variants (".en") tend to perform better than the multilingual checkpoints for English applications, particularly for the base size. On the LibriSpeech test-clean benchmark, whisper-base.en achieves a Word Error Rate (WER) of 4.27%. On the more challenging LibriSpeech test-other set, it achieves a WER of 12.80%. On the Open ASR Leaderboard aggregated benchmark, the model reports a mean WER of 10.32 across datasets including AMI (21.13%), Earnings22 (15.09%), and GigaSpeech (12.83%). Key capabilities include: - Robustness to accents, background noise, and technical language - Zero-shot transfer to new domains without fine-tuning - Long-form transcription via chunking (up to arbitrary audio length) - Timestamp prediction for transcribed segments The model has 74 million parameters and is designed for audio samples up to 30 seconds in duration. For longer recordings, chunking with batched inference is supported. The model is hosted as a managed API on gigarouter, providing OpenAI-compatible endpoints for direct integration.