Whisper (whisper-small)

OpenAI's 244M parameter transformer model for high speed multilingual speech recognition and translation.

Whisper-small strikes the optimal balance between computational efficiency and transcription accuracy. This 244 million parameter model (trained on 680,000 hours of labeled audio) handles multilingual speech recognition and translation across 99 languages. It operates significantly faster than the large-v3 variant while maintaining low Word Error Rates (WER) on standard benchmarks like LibriSpeech. Developers integrate this checkpoint via the Hugging Face Transformers library to power real time captioning or voice interfaces. It excels at identifying technical terminology and filtering background noise without requiring extensive fine tuning.

https://huggingface.co/openai/whisper-small