Alibaba Cuts AI Audio Prices by Up to 95%

Alibaba announces significant price reductions on its AI audio services, with discounts of up to 95% for speech recognition. At the same time, Qwen-Audio-3.1 introduces five new models covering recognition, synthesis, and real-time interaction, with extended features for ASR, TTS, and voice agents.

Falling Prices for TTS, Real-Time, and ASR

Alibaba is slashing prices for its AI audio services. The rates for text-to-speech (TTS) are decreasing by about 70%, real-time interaction capabilities by about 85%, and speech recognition (ASR) by up to 95%. These reductions apply to the TTS, real-time, and ASR categories.

Speech Recognition: Multilingual, Dialects, and Sound Context

The ASR model of Qwen-Audio-3.1 enhances support for multiple languages and various dialects while automatically eliminating filler words and repetitions. The ASR-Next version also allows for the identification of multiple speakers with time markers and recognizes emotions, background noise, as well as machine sounds.

Synthesis and Interaction: One-Pass Streaming and Tailored Responses

For text-to-speech, Qwen offers multilingual generation with natural voice transfer between languages and allows control over emotion, speed, and style through simple text instructions. TTS-Next combines a linguistic model and a streaming method to simultaneously generate voice, sound effects, and audio background in a single step. The real-time model permits speaking and listening at the same time, with the ability to interrupt instantly. According to Qwen, low mood detection results in slower responses and increased empathy. Qwen-Audio-3.1 thus consolidates five models covering these uses, with additional information available on the blog and Qwen Cloud.