Automatic Speech Recognition (ASR) with llama.cpp and Qwen3-ASR
- 2 minutes read - 301 wordsAutomatic Speech Recognition (ASR) has become an essential capability for many AI-powered applications. Whether it is transcribing meetings, generating subtitles, or enabling voice-based interactions, ASR is increasingly becoming a core feature of modern AI systems.
As part of my ongoing learning journey with AI technologies, I decided to start exploring the ASR capabilities available today. My first stop is Qwen3-ASR, one of the latest speech recognition models from the Qwen family.
Why Qwen3-ASR?
There are many open-source ASR models available today, but Qwen3-ASR caught my attention because it extends the Qwen ecosystem into speech understanding. Since I have already been experimenting with various LLMs and inference engines, it felt natural to begin my ASR exploration with this model.
Rather than diving straight into benchmarks, I wanted to understand the fundamentals:
How to run the model locally What the inference workflow looks like How well it performs on real audio samples How easy it is to integrate into future AI applications
This exploration is only the beginning, but it provides a solid foundation for understanding modern speech recognition models.
Recording the Journey
To document the setup process, I recorded the entire terminal session using Asciinema.
Unfortunately, Asciinema is currently unavailable, so I cannot host the recording there.
Instead, I will publish the recording on my own blog:
Hopefully this makes the walkthrough available to anyone interested in reproducing the setup.
What’s Next?
This is the first article in what I expect will become a small series on ASR.
Some topics I plan to explore include:
Running Qwen3-ASR with different inference backends Comparing transcription quality with other open-source ASR models Performance optimization Integrating ASR into AI agents and voice-enabled applications Building practical speech-to-text workflows
Stay tuned as I continue exploring the rapidly evolving world of open-source speech recognition.