Voice AI Weekly #8 - First Week of September 2026
In the global voice AI industry this week, the race for new speech recognition models and contact center automation took center stage. At the same time, the debate over voice security and ownership has entered a new phase.
1. Google Unveils Preview of Gemini 3.5 Transcribe, a New Speech-to-Text Model
On August 26, Google launched the public preview of its speech recognition model 'Gemini 3.5 Transcribe'. Supporting more than 85 languages, Google reports an average Word Error Rate (WER) of 2.6% for audio files and 4.0% for real-time streaming. It includes a feature that cleans up background noise, jargon, and even stutters, converting them into clean text.
While general-purpose models are expanding their reach into speech habits and terminology processing, it remains to be seen whether they can effectively capture specialized industry terms or regional dialects. This ultimately highlights once again why domain-specific fine-tuning is necessary.
Source: Intelligent transcription with Gemini 3.5 Transcribe
2. Australia's Apate.AI Raises Investment for Adversarial Voice Agents Targeting Scammers
Australian startup Apate.AI raised AUD 11.4 million in seed funding on August 31. The company's voice and text agents act like real victims, holding conversations with scammers to waste their time and extract information. Commonwealth Bank of Australia stated that they have used this agent to conduct over 2.5 million automated conversations, securing more than 250,000 pieces of intelligence.
Wasting scammers' time and extracting information by putting forward fake victim voices shows a shift from mere defense to proactive response. In the voice security domain, the role of voice AI is growing significantly.
Source: Apate.AI raises $11.4M Seed to turn AI scammers against themselves
3. Natera Reveals Amazon Bedrock-Based Voice Scheduling Agent
On August 26, US genetic testing company Natera unveiled a voice agent powered by Amazon Bedrock AgentCore. Designed for patients to schedule mobile blood draw appointments over the phone, the service achieved 100% tool-calling accuracy and a response latency of under 7 seconds.
Voice agents are beginning to be deployed in real customer service lines starting with scheduling tasks, which are repetitive but require high accuracy. As real-world use cases accumulate in the healthcare sector, the barrier to adoption in the field is naturally expected to lower.
Source: Natera's intelligent appointment scheduling with Amazon Bedrock AgentCore
4. Liveops Launches LiveNexus, a Real-Time Agent Assistance Tool
On August 27, Liveops, a US customer experience outsourcing company, launched LiveNexus Agent Assist, a real-time agent assistance tool. Layered as a browser overlay on top of existing CRM and contact center systems, it automatically recommends next steps during calls, guides agents through compliance procedures, and generates post-call summaries.
Adding intelligence on top of existing workflows without moving agents to a completely new system is a practical approach that lowers barriers to adoption. As call data is automatically accumulated, it seamlessly connects to subsequent quality control and compliance audits.
Source: Liveops Launches LiveNexus Agent Assist
5. Over 80 British Actors Call for Legislation to Ban AI Voice Cloning
On August 28, over 80 British actors, including Nicola Coughlan and Hugh Bonneville, joined a campaign demanding legislation to block AI voice cloning. They pointed out that a voice can be cloned with just three seconds of audio, and reported that 28% of UK adults have been targeted by voice cloning scams.
The discussion that voice should be treated as personal property is gaining momentum outside the industry. For companies handling voice data, establishing consent procedures and protection safeguards beforehand is the safest path forward.
Source: Nicola Coughlan and Hugh Bonneville among stars backing campaign against AI voice cloning
While the race for speech recognition models and contact center automation accelerates, the issue of protecting voices is also gaining significant weight.
Please feel free to reach out to contact@holamago.com at any time for inquiries or collaboration proposals. We will return next week with more voice AI news curated from MAGO's perspective.