wames123/airgrep-asr-gemma-4-e4b
wames123/airgrep-asr-gemma-4-e4b is a 7.9 billion parameter Gemma-4-e4b LoRA fine-tune, specifically designed to enhance Automatic Speech Recognition (ASR) performance on radio transmissions. This model integrates an audio tower to improve its capability in processing and transcribing radio communication. Its primary application is in specialized ASR tasks, particularly for environments involving radio frequency audio.
Loading preview...
Model Overview
wames123/airgrep-asr-gemma-4-e4b is a specialized 7.9 billion parameter model, developed as a LoRA fine-tune of the Gemma-4-e4b architecture. This fine-tuning specifically targets and enhances the model's audio tower capabilities.
Key Capabilities
- Enhanced ASR for Radio Transmissions: The model is engineered to perform Automatic Speech Recognition (ASR) with improved accuracy on audio derived from radio communications.
- Gemma-4-e4b Base: Leverages the foundational strengths of the Gemma-4-e4b model, extended with a focus on audio processing.
- LoRA Fine-tuning: Utilizes Low-Rank Adaptation (LoRA) for efficient and targeted fine-tuning, allowing for specialized performance without extensive retraining of the base model.
Use Cases
- Radio Communication Transcription: Ideal for transcribing speech from various radio transmission sources.
- Specialized ASR Applications: Suitable for scenarios where standard ASR models may struggle with the unique characteristics of radio audio.
This model was developed as part of the Gemma 4 Good Kaggle Hackathon, with the broader AirGrep project available on GitHub.