Whisper-NER v1
| Property | Value |
|---|---|
| Parameter Count | 1.54B |
| License | MIT |
| Paper | WhisperNER Paper |
| Tensor Type | F32 |
What is whisper-ner-v1?
Whisper-NER v1 is an innovative model that combines automatic speech recognition (ASR) with named entity recognition (NER) in a unified architecture. Developed by aiola, this model represents a significant advancement in joint speech transcription and entity recognition tasks, particularly notable for its support of open-type NER capabilities.
Implementation Details
The model is built upon the Whisper architecture and has been specifically trained on the NuNER dataset for English language processing. It utilizes a transformer-based architecture with 1.54 billion parameters, implementing F32 tensor types for computation. The model can be easily integrated using the Hugging Face Transformers library and supports custom entity tag prompting during inference.
- Trained specifically on English language data
- Supports open-type named entity recognition
- Implements prompt-based entity specification
- Operates at 16kHz sample rate for audio input
Core Capabilities
- Joint speech transcription and entity recognition
- Real-time processing of audio inputs
- Flexible entity type recognition through prompting
- Support for various audio input formats
- Efficient processing with GPU acceleration support
Frequently Asked Questions
Q: What makes this model unique?
This model stands out for its ability to perform both speech recognition and named entity recognition simultaneously, supporting open-type NER that allows for recognition of diverse and evolving entities without being constrained to predefined categories.
Q: What are the recommended use cases?
The model is ideal for applications requiring automatic transcription with entity extraction, such as content analysis, automated meeting transcription, and information extraction from audio sources. It's particularly useful when specific entity types need to be identified within spoken content.
