Qwen3-ASR Technical Report

TL;DR

The paper tackles robust multilingual automatic speech recognition with precise timestamp alignment in a unified, all-in-one framework. It introduces two ASR models (Qwen3-ASR-1.7B and Qwen3-ASR-0.6B) and a non-autoregressive forced aligner (Qwen3-ForcedAligner-0.6B) built on the Qwen3-Omni foundation and an AuT encoder, with a multi-stage training pipeline (AuT pretraining, Omni pretraining, ASR fine-tuning, and RL). The approach achieves state-of-the-art open-source ASR performance, strong language identification, singing voice transcription, and efficient streaming inference, while the forced aligner provides fast, accurate, multilingual timestamp predictions across 11 languages. The models are open-sourced under the Apache 2.0 license, offering a practical, scalable toolkit for research and deployment that advances speech understanding and timestamping in real-world scenarios. Overall, Qwen3-ASR demonstrates robust multilingual coverage, competitive performance against proprietary APIs, and a novel timestamping capability that fills a critical gap in the speech technology stack.

Abstract

In this report, we introduce Qwen3-ASR family, which includes two powerful all-in-one speech recognition models and a novel non-autoregressive speech forced alignment model. Qwen3-ASR-1.7B and Qwen3-ASR-0.6B are ASR models that support language identification and ASR for 52 languages and dialects. Both of them leverage large-scale speech training data and the strong audio understanding ability of their foundation model Qwen3-Omni. We conduct comprehensive internal evaluation besides the open-sourced benchmarks as ASR models might differ little on open-sourced benchmark scores but exhibit significant quality differences in real-world scenarios. The experiments reveal that the 1.7B version achieves SOTA performance among open-sourced ASR models and is competitive with the strongest proprietary APIs while the 0.6B version offers the best accuracy-efficiency trade-off. Qwen3-ASR-0.6B can achieve an average TTFT as low as 92ms and transcribe 2000 seconds speech in 1 second at a concurrency of 128. Qwen3-ForcedAligner-0.6B is an LLM based NAR timestamp predictor that is able to align text-speech pairs in 11 languages. Timestamp accuracy experiments show that the proposed model outperforms the three strongest force alignment models and takes more advantages in efficiency and versatility. To further accelerate the community research of ASR and audio understanding, we release these models under the Apache 2.0 license.

Figures (3)

• Figure 1: Qwen3-ASR family includes all-in-one ASR models with advantages in multilingual, noisy speech recognition, singing voice recognition and inference efficiency, so as to a novel multilingual speech forced alignment model for predicting timestamps of words or sentences in ASR results.
• Figure 2: Architecture of AuT (left) and the overview of Qwen3-ASR (right).
• Figure 3: Illustration of Qwen3-ForcedAligner-0.6B. During training, randomly masked timestamp slots with are dynamically inserted into the token sequence to represent word or character boundaries. The combined sequence is fed into Qwen3-0.6B LLM, and a timestamp prediction layer predicts the corresponding timestamp indices for each slot. Supervision is applied with cross‑entropy loss on synchronously aligned label and output sequences.