AI Voice Recognition Slashed by Up to 95%——Alibaba's Qwen-Audio 3.1 Shakes the Industry
Alibaba's 'Qwen-Audio-3.1,' announced on September 23, 2026, delivers a massive price reduction: 95% off speech recognition (ASR) and 70% off text-to-speech (TTS). The era of near-unlimited AI voice capabilities at almost no cost has arrived.
On September 23, 2026, at the 2026 Apsara Conference held in Hangzhou, China, Alibaba announced its voice AI model "Qwen-Audio 3.1." With cuts of up to 95% on speech recognition (ASR), 85% on real-time voice conversation, and 70% on text-to-speech (TTS), the AI voice market has been sent into shock.
What Is Qwen-Audio 3.1? (Alibaba's AI Voice Model)
"Qwen" is a series of AI models developed by Alibaba. The voice-specialized edition is "Qwen-Audio," and version 3.1 brings a major expansion of features alongside sweeping price reductions.
Qwen-Audio 3.1 offers three main categories of voice processing:
- ASR (Automatic Speech Recognition): Converts spoken language into text. Think of it as automatically transcribing a recorded meeting.
- TTS (Text-to-Speech): Converts written text into human-like speech. Used for auto-generating narration and voice announcements.
- Realtime: Enables near-instantaneous voice conversation between humans and AI.
How Much Did Prices Drop? (Concrete Comparison)
The scale of this price cut is unprecedented in the industry.
Price Cut Summary
| Model | Function | Price Reduction |
|---|---|---|
| ASR | Speech → Text conversion | -95% |
| Realtime | Real-time voice conversation | -85% |
| TTS | Text → Speech synthesis | -70% |
| ASR-Next | Voice understanding, summarization, sentiment analysis | New (vs. previous gen) |
| TTS-Next | Voice + sound effects + BGM all-in-one generation | New (vs. previous gen) |
What You Can Do with the 5 Models
Qwen-Audio 3.1 consists of five distinct models. Here is a breakdown of each model's features and practical use cases for everyday life and business.
1. ASR (Speech Recognition)
Converts spoken language into text in real time. This is the model receiving the 95% price cut. It can be used for auto-transcribing podcasts, generating meeting minutes, and analyzing call center recordings.
2. TTS (Text-to-Speech)
Converts written text into natural-sounding speech. With a 70% price reduction, applications include audio editions of news articles, read-aloud features in language learning apps, and automated announcement generation — all at a fraction of the previous cost.
3. Realtime (Real-Time Voice Conversation)
Enables nearly lag-free conversation with an AI. An 85% price cut makes it feasible to build voice-capable AI customer support systems and language practice bots at realistic costs.
4. ASR-Next (New)
A new model that goes beyond simply converting speech to text — it specializes in voice "understanding." In a single API call, it can summarize meeting recordings, perform sentiment analysis on individual speakers, and identify who is speaking when multiple people are talking.
5. TTS-Next (New)
A next-generation voice synthesis model combining language models and diffusion models. It generates not just voice but entire soundscapes — including sound effects and BGM — in a single API call. Tasks like film narration or advertising audio production, which previously required multiple tools, are dramatically simplified.
Multilingual and Dialect Support
Qwen-Audio 3.1 supports multilingual text-to-speech including regional dialects. Its "cross-lingual voice transfer" feature can even preserve the vocal characteristics of an English speaker while producing speech in Japanese — switching the language without changing the voice identity.
How Does This Affect Our Daily Lives and Work?
These price cuts are not just a story for developers. The impact will gradually ripple into everyday life and the workplace.
Podcasters and video creators: Dramatically lower transcription costs make it far easier to generate subtitles and search through content.
Call centers and customer support: The barrier to adopting AI for call analysis, quality evaluation, and automated complaint categorization falls significantly.
Language learning apps: High-quality TTS at lower costs means more natural learning experiences can be delivered affordably.
Meeting and interview management: Lower costs for real-time transcription and summarization make these tools accessible even to small and mid-sized businesses.
Ripple Effects on the AI Voice Price War
The announcement of Qwen-Audio 3.1 could trigger pricing pressure across the voice features of ChatGPT (OpenAI), Gemini (Google), and Claude (Anthropic).
OpenAI's Whisper (speech recognition) and TTS APIs are already competitively priced, but Alibaba's 95% price cut puts pressure on all players to follow suit. The practical impact: the cost of deploying AI voice features in Japanese apps, games, and call centers is expected to drop significantly in the near future.
Summary
Alibaba's Qwen-Audio 3.1 has the potential to fundamentally reshape the pricing landscape for voice AI. The 95% ASR price cut is not just a promotional discount — it signals that the era of voice AI accessible to everyone has arrived.
Going forward, voice capabilities like transcription, speech synthesis, emotion analysis, and multilingual support will become standard features embedded in smartphone apps and enterprise systems. This announcement may well be remembered as the turning point when voice AI shifted from a "special feature" to everyday infrastructure.
Related Articles
More Llama articles
Muse Charm vs. Ray-Ban Meta Gen 3 vs. Ray-Ban Audio: Which Meta Device Should You Choose?
A deep-dive comparison of the three devices announced at Meta Connect 2026. We break down the differences between the camera-free music-focused Ray-Ban Meta Audio ($349), the full-featured AI glasses Ray-Ban Meta Gen 3 ($449), and the AI keychain Muse Charm (December, price TBD) — and help you pick the right one.
Meta Unveils 'AI Keychain' — Muse Charm Lets You Talk to AI Right from Your Key Ring
At Meta Connect 2026 on September 23, 2026, Meta announced the keychain-sized AI gadget 'Muse Charm' (launching December), lightweight VR glasses ($1,299, Spring 2027), and Ray-Ban Meta Gen 3 ($449, available now). The era of standalone AI hardware that works without a smartphone has officially arrived.
Talk to an AI Without Touching Your Phone — Meta × Ray-Ban Smart Glasses Now Available in Japan
On May 21, 2026, Ray-Ban Meta (Gen 2) officially launched in Japan starting at ¥73,700. Just say 'Hey Meta' and the AI responds — point the camera at a foreign-language sign and it translates to Japanese, or capture photos and videos completely hands-free. A new way to interact with AI without ever taking your phone out of your pocket.