WhisperKit not recording filler words

[zaidbren]

Hello everyone, I am using WhisperKit speech to text in order to build transcription based editing feature, and I need to perfectly get all the filler words during the recording. I used whisperKit large model for speech to text, however, it doesn't give filler words like umms, ahhs, and many more, it just give perfect transcription without the actual spoken words :-

let options = DecodingOptions(
            task: .transcribe,
            language: selectedLanguage.code,
            skipSpecialTokens: true,
            wordTimestamps: true,
            supressTokens: nil
        )

For example, in this sentence, I spoken a lot of "ums" and "ahs" however they are not being given :- "However, I can see that there's a lot of the things that I want to point out exactly"

How can I get accurate filler words, with actual words spoken etc.?

[atiorh]

@zaidbren This is related to the model behavior and not tied to WhisperKit. WhisperKit does not filter filler words. There are other models that transcribe filler words such as Nvidia Parakeet v2. However, that model is not supported in WhisperKit yet. It is available in the Argmax Pro SDK and in various other open-source projects.