88Z AI, 1 September — Xiaohongshu's FireRed team has released FireRedAudio, a general audio language model based on the Qwen3.5 architecture. It supports six major tasks including ASR and audio QA, achieving leading results in multilingual evaluations.
At a glance
- Built on Qwen3.5 architecture, supporting six audio tasks including ASR and TTS
- Achieved a 14.94% average error rate on FLEURS-102, leading in multilingual evaluations
The update
The FireRed team at Xiaohongshu has officially launched FireRedAudio, a general-purpose audio language model. Built on the Qwen3.5 architecture with 9 billion parameters, it utilizes a decoupled continuous representation design. Functionally, FireRedAudio handles six core tasks simultaneously: automatic speech recognition, audio question answering, long-form audio understanding of up to one hour, zero-shot text-to-speech, instruction-based text-to-speech, and voice editing. The model has achieved leading results in various authoritative benchmarks, including MMAU, MMSU, automatic speech recognition across 102 languages, and instruction-based text-to-speech tests. Notably, it recorded an average error rate of just 14.94% on the FLEURS-102 benchmark, demonstrating robust multilingual processing capabilities and advanced audio understanding.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- FireRedAudioSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.