Trellis Data's Egyptian Arabic model has been fine-tuned on 120 hours of audio recordings. Our testing shows 38% WER; we are confident that this could be improved further with additional data. This model is available for usage as-is through Amazon SageMaker. Please reach out if you'd like to work together to develop further improvements.

Trellis Data addresses the limitations of off-the-shelf translation and transcription tools by developing custom AI solutions capable of understanding and processing bespoke languages. We create production ready models that bridge the gap where mainstream options fall short, equipping enterprises, government agencies, and high-security organizations with language solutions tailored to their unique needs and use cases. We specialize in training, fine-tuning, and optimizing speech-to-text AI models, ensuring accurate output for niche, low-resource, or specialized languages. Our models are designed to deliver optimal results, while providing the linguistic accuracy and cultural sensitivity required for mission-critical applications. Our models are already deployed and trusted by law enforcement and high security agencies in Government.