Compare Vocola 3 vs. Whisper

Windows Speech Recognition (WSR) performs effectively in applications that are compatible with it, such as MS Word, Outlook, and PowerPoint, allowing for seamless dictation where text is inserted directly into documents and commands like "Delete hedgehog" target specific text. However, in applications that are not optimized for WSR, including MS Excel, Gmail, and various programming environments, dictation struggles, as the spoken words do not integrate into the document text, and commands lack the capability to refer to existing document content. Vocola addresses these limitations by enabling direct dictation in WSR-unfriendly applications and facilitating the correction and alteration of the most recently spoken phrase. Both Vocola and WSR utilize the same speech profile, meaning that any enhancements from training, corrections, or adjustments to the speech dictionary will improve dictation capabilities in both systems equally. Unfortunately, on the Vista operating system, dictation in non-friendly applications is particularly problematic, as every spoken command triggers the correction panel, rendering the feature nearly ineffective. Overall, while WSR is beneficial for compatible applications, the experience can be significantly hindered when trying to use it in others.

We have developed and are releasing an open-source neural network named Whisper, which achieves levels of accuracy and resilience in English speech recognition that are comparable to human performance. This automatic speech recognition (ASR) system is trained on an extensive dataset comprising 680,000 hours of multilingual and multitask supervised information gathered from online sources. Our research demonstrates that leveraging such a comprehensive and varied dataset significantly enhances the system's capability to handle different accents, ambient noise, and specialized terminology. Additionally, Whisper facilitates transcription across various languages and provides translation into English from those languages. We are making available both the models and the inference code to support the development of practical applications and to encourage further exploration in the field of robust speech processing. The architecture of Whisper follows a straightforward end-to-end design, utilizing an encoder-decoder Transformer framework. The process begins with dividing the input audio into 30-second segments, which are then transformed into log-Mel spectrograms before being input into the encoder. By making this technology accessible, we aim to foster innovation in speech recognition technologies.