Alibaba's New AI Cuts Audio Costs by 95% and Adds Human-Like Voice

Alibaba's Qwen team just launched Qwen-Audio-3.1, a collection of five AI models that handle different audio tasks. The most striking change is the price: using their text-to-speech service now costs about 70% less, their real-time conversation service is 85% cheaper, and their speech recognition service is up to 95% cheaper. That means businesses and developers can build voice-powered apps for a fraction of what it used to cost.

The new models are also smarter. The speech recognition model can understand multiple languages and dialects, and it automatically removes filler words like "um" and "uh" from transcripts. Another model can identify different speakers in a conversation and even detect emotions and background sounds. The text-to-speech model lets you control the emotion, speed, and style of the voice with simple instructions.

There's also a model called TTS-Next that can generate voices, sound effects, and background audio all at once—useful for creating audiobooks, podcasts, or game audio. And the real-time model lets you have a natural back-and-forth conversation with the AI, including interrupting it, just like a phone call. It can even sense if you're feeling down and respond with empathy.

For everyday people, this means voice assistants and automated customer service will become more natural and responsive. Content creators can produce professional-sounding audio cheaply. And with the huge price cuts, we'll likely see more apps and services that use voice AI, making it a normal part of daily life.