Alibaba's Qwen-Audio-3.1 cuts audio AI prices up to 95 percent

Alibaba's Qwen team has expanded its audio model lineup with Qwen-Audio-3.1, a set of five models spanning speech recognition (ASR), text-to-speech (TTS), and real-time interaction. The release includes improvements such as multilingual and dialect recognition in the ASR model, automatic cleanup of filler words, and new capabilities in ASR-Next for multi-speaker identification with timestamps and detection of emotions, ambient sounds, and machine noise. TTS now handles multilingual synthesis with natural cross-language voice transfer, and users can control emotion, speed, and style through simple text prompts.

The lineup also introduces TTS-Next, which pairs a language model with a diffusion approach to generate voice, sound effects, and background audio in a single pass, and a real-time model that supports simultaneous speaking and listening with instant interruption. According to Qwen, when the real-time model detects a low mood, it responds more slowly and with more empathy.

Alongside the release, Alibaba is cutting prices. TTS drops about 70 percent, Realtime roughly 85 percent, and ASR up to 95 percent. The combination of a broader, more capable lineup and steeper discounts could lower the cost for developers building speech-based applications. The outcome hinges on whether the models perform reliably across diverse languages and audio conditions in real-world use, and how quickly developers adopt the cheaper options. More details are available on the blog and on Qwen Cloud.