Fact · Voice and AI
NVIDIA's Canary-Qwen model maps the audio encoder's output into the language model's embedding space through a linear projection.
Checked against its source on October 17, 2026
In the source's own words
“The audio encoder computes audio representation that is mapped to the LLM embedding space via a linear projection”
canary-qwen-2.5b model card, NVIDIA, Hugging Face, n.d., Model Architecture.
Open the original- What it measures
- The model card says the design pairs a FastConformer audio encoder with the Qwen3-1.7B language model, and that low-rank adaptation (LoRA) is applied to the language model.
- What it does not show
- This describes the model's design from its model card. The benchmark numbers for the model come from the leaderboard, not from the card.