Qwen Releases Compact ASR Model for Streaming Audio
The new Fun-ASR-Nano model from Alibaba's team packs real-time multilingual transcription, speaker diarization, and hotword detection into an efficient package.
Alibaba's Qwen team has released Fun-ASR-Nano-2512, a new automatic speech recognition (ASR) model designed for efficiency and real-time performance. As its "Nano" designation suggests, the model is compact, making it a candidate for applications where computational resources are constrained.
Fun-ASR-Nano moves beyond simple transcription by integrating several advanced features often found in much larger systems. Its architecture is built for streaming audio, allowing it to process speech with low latency as it's spoken, rather than waiting for an entire audio file to be complete.
Structured Audio Output
This combination of features makes the model particularly useful for building sophisticated conversational AI and analysis tools. Key capabilities detailed in the official release include:
- Speaker diarization: Identifying who is speaking and when.
- Word-level timestamps: Aligning transcribed text with its precise timing in the source audio.
- Hotword detection: Customizing the model to reliably recognize specific keywords.
- Multilingual support: Processing speech from multiple languages.
By packaging these tools into a lightweight model, the Qwen team provides a powerful component for developers creating on-device or edge applications, such as smart meeting assistants or embedded voice-controlled interfaces. The model is available under the custom Model-Scope Open-Source License.
Sources
- Visit
FunAudioLLM/Fun-ASR-Nano-2512
Hugging Face
0 comments
No comments yet. Be the first to weigh in.
More in Speech → Text
Mega-ASR Improves on Qwen for Speech Recognition
Researcher Zhifei Xie has released a 1.7B-parameter model that refines Alibaba's Qwen3-ASR, showing improved performance on English and Chinese transcription benchmarks.
NVIDIA Releases Nemotron-3.5 Streaming ASR Model
The 600-million-parameter model uses a FastConformer architecture for real-time, multilingual speech-to-text applications.
Xiaomi Releases MiMo Model for Speech Recognition
The new open-source model from the Chinese tech giant offers automatic speech recognition for Mandarin, Cantonese, and English under a permissive MIT license.