CrisperWhisper

Word-level automatic speech recognition model

CommonProductProductivityAutomatic Speech RecognitionVerbatim Transcription

CrisperWhisper is an advanced variant of OpenAI's Whisper model, specifically designed for fast, accurate, verbatim speech recognition, providing precise word-level timestamps. Unlike the original Whisper model, CrisperWhisper aims to transcribe every spoken word, including filler words, pauses, stutters, and false starts. This model ranks first in word-level datasets such as TED and AMI, and has been accepted at INTERSPEECH 2024.

Accurate word-level timestamps: Provides precise timestamps even at points of disfluency and pauses.
Verbatim transcription: Includes every word
distinguishing filler words like 'um' and 'uh'.
Filler word detection: Identifies and accurately transcribes filler words.
Illusion reduction: Minimizes transcription hallucinations to enhance accuracy.
Support for streaming applications: Offers a user-friendly interface via the Streamlit application
allowing users to record or upload audio files for transcription.
High performance: Significantly outperforms Whisper Large v3 across multiple datasets
particularly in verbatim transcription styles.

CrisperWhisper is ideal for researchers and developers who require high-precision speech recognition
especially in scenarios that demand verbatim transcription and analysis of spoken language
such as meeting minutes
lecture transcriptions
and language learning.

Researchers use the CrisperWhisper model to analyze speech patterns in TED talks.
Educational institutions leverage this model to enhance the transcription quality of language learning materials.
Companies use CrisperWhisper to automatically generate meeting minutes and summaries.

1. Clone the CrisperWhisper repository to your local machine.
2. Create and activate a Python virtual environment.
3. Install the necessary dependencies.
4. Download the model using a Hugging Face account.
5. Utilize the model for speech recognition through a Python script or Streamlit application.

Visit