Alibaba has released the Qwen-Audio-3.1 series, a collection of five new audio models covering understanding, generation, interaction, and creation. The update includes the Qwen-Audio-3.1-TTS-Next for audio creation and the Qwen-Audio-3.1-ASR-Next for audio understanding.
The company has significantly reduced prices across its entire Qwen-Audio product line. TTS services saw a price cut of approximately 70%, Realtime services were reduced by about 85%, and ASR services saw a reduction of 95%.
The Qwen-Audio-3.1-ASR model supports 30 languages and 16 Chinese dialects. It features native transcription polishing, speaker diarization, and low-latency streaming with a first-character response time of approximately 160 milliseconds.
The Qwen-Audio-3.1-TTS model enables unified generation of human voice, sound effects, and ambient noise. It supports multi-speaker cloning, emotional control, and 48kHz output, catering to professional needs in podcasting, gaming, and film production.
The Qwen-Audio-3.1-Realtime model introduces full-duplex interaction, allowing users to interrupt and speak simultaneously. It can detect emotions, maintain context during language switching, and call external tools or APIs to execute tasks.
These models are being integrated into various hardware and software solutions, including Qoder, Qwen Office, QwenNote, and AI glasses from Leqi and Eva, allowing users to initiate tasks directly via voice commands.
