Whisper

Whisper, created by OpenAI, is a versatile speech recognition model designed for a variety of audio applications. It has been trained on an extensive dataset featuring diverse audio inputs and functions as a multi-task model capable of multilingual speech recognition, speech translation, and language detection. Utilizing a Transformer sequence-to-sequence architecture, Whisper addresses several speech processing challenges, such as multilingual recognition, spoken language identification, and voice activity detection. By representing these tasks as a series of tokens for the decoder to predict, Whisper streamlines the traditional speech-processing workflow into a single model. Its multitask training incorporates special tokens that act as task identifiers or classification targets.

Voice & Transcription