Qwen Cuts Qwen-Audio Voice Model Prices; ASR Down 95%

On September 23, Qwen announced price cuts across its Qwen-Audio voice model lineup, with TTS prices reduced by about 70%, Realtime by about 85%, and ASR by 95%, aiming to lower the cost for users. The company also released the Qwen-Audio-3.1 series of voice models on the same day, covering speech recognition, speech synthesis, real-time voice interaction, and audio creation.

The Qwen-Audio-3.1-ASR is a new-generation speech recognition model supporting 30 languages and 16 Chinese dialects. It adds native transcription refinement, removing filler words and repeated expressions, and supports speaker-attributed transcription. The Qwen-Audio-3.1-ASR-Next is an audio understanding model built on a next-generation architecture, capable of recognizing emotions, ambient sounds, and mechanical sounds, as well as answering questions about audio content.

The Qwen-Audio-3.1-TTS supports multilingual and dialect synthesis with natural cross-language voice transfer. The Qwen-Audio-3.1-TTS-Next can generate speech, sound effects, and ambient audio from text, timestamps, and reference audio in one pass, for use in podcasts, audiobooks, film, TV, and games.

The Qwen-Audio-3.1-Realtime supports full-duplex interaction, allowing simultaneous listening and speaking, with interruptions, and can call Agent tools to complete tasks. It is being integrated with agents including Qoder and Qwen Office, as well as smart hardware such as QwenNote, A2, Eva, and Qwen AI Glasses.

According to Qwen, the five new voice models form a complete audio capability stack covering understanding, generation, interaction, and creation.