Alibaba rolls out Qwen Audio 3.1 alongside new models, cutting AI audio costs by as much as 95%

Alibaba's AI research team Qwen has unveiled Qwen-Audio-3.1, a family of five models covering speech recognition, text-to-speech, and real-time voice interaction. Alongside the release, the company is cutting prices across its audio services by between 70 and 95 percent, signaling an aggressive push into the voice AI market.

Smarter Speech Recognition

The upgraded ASR model delivers stronger multilingual and dialect performance and automatically strips out filler words and repetitions from transcriptions. Its companion model, ASR-Next, goes further by identifying multiple speakers with timestamps, while also detecting emotions, ambient sounds, and machine noise in recordings.

Natural Text-to-Speech

Qwen's TTS model supports multilingual synthesis with smooth voice transfer across languages. Users can steer delivery through plain-text instructions — for example, "Read this with a sharp, commanding tone, demanding respect" — controlling emotion, pace, and speaking style.

TTS-Next combines a language model with a diffusion-based approach to generate voice, sound effects, and background audio together in a single pass.

Real-Time Conversation

The real-time model enables simultaneous speaking and listening with instant interruption support. Qwen says it can also pick up on a user's low mood, responding more slowly and with greater empathy.

Steep Price Cuts

Alibaba is reducing costs significantly: TTS pricing drops roughly 70 percent, Realtime falls by about 85 percent, and ASR sees reductions of up to 95 percent. Full details are available on the Qwen blog and Qwen Cloud. The move positions Qwen to compete for developers building voice-first applications, and pricing responses from rival providers will be worth watching.