Qwen3-ASR
An automatic speech recognition system developed by Alibaba based on the Qwen large language model. It follows an LLM-based ASR architecture and supports multilingual speech recognition tasks.
Core Facts
Timeline (last 90 days)
Pretrained ASR systems (such as Whisper and Qwen3-ASR) perform well on standard corpora but perform poorly in high-noise domains such as broadcast police communications.
The paper proposes a cross-model pseudo-labeling paradigm: using pseudo-labels generated by one model to fine-tune another model.
The two foundation models involved in the study are Whisper and Qwen3-ASR.
Nari has launched two speech models based on Qwen3: a text-to-speech (TTS) model and an automatic speech recognition (ASR) model
The study employs three ASR architectures: MMS (encoder-only), Whisper (encoder-decoder), and Qwen3-ASR (LLM-based).
Alibaba's Qwen team released the speech recognition model Qwen-Audio-3.0-ASR-Flash.
All Facts (6)
Alibaba's Qwen team released the speech recognition model Qwen-Audio-3.0-ASR-Flash.
65%UnverifiedPretrained ASR systems (such as Whisper and Qwen3-ASR) perform well on standard corpora but perform poorly in high-noise domains such as broadcast police communications.
50%UnverifiedThe paper proposes a cross-model pseudo-labeling paradigm: using pseudo-labels generated by one model to fine-tune another model.
50%UnverifiedThe two foundation models involved in the study are Whisper and Qwen3-ASR.
50%UnverifiedThe study employs three ASR architectures: MMS (encoder-only), Whisper (encoder-decoder), and Qwen3-ASR (LLM-based).
50%UnverifiedNari has launched two speech models based on Qwen3: a text-to-speech (TTS) model and an automatic speech recognition (ASR) model
60%