Voice AI Weekly #5 - 2nd Week of August 2026
We have gathered five noteworthy news items from the overseas voice AI industry over the past week. We look at the trends, from regulations to corporate earnings and technology announcements.
1. ElevenLabs Releases Emotion-Preserving Dubbing Engine as API
On August 6th, ElevenLabs opened its dubbing engine, Dubbing v2, as an API for developers. Supporting over 90 languages, it bypasses text translation to directly extract intonation, speech rate, and emotional expression from the original audio signal and apply them to the dubbing. Initially released as a consumer feature last May, developers can now integrate it into their own services. Major companies are also joining this trend of handling emotions and intonations prior to the text translation stage.
Source ElevenLabs Opens Dubbing API: Emotion-Preserving AI Localization Now Programmable
2. xAI Officially Transitions to Speech-to-Speech Model Grok Voice Think Fast 2.0
Grok Voice Think Fast 2.0, the speech-to-speech model released by xAI on July 29th, became the default API model on August 5th. Bypassing the text translation step to listen and respond directly, it reduced the time to the first voice response from 1.25 seconds to 0.70 seconds. In a 24-language evaluation, it reportedly recorded 1.5 to 2 times higher accuracy compared to Deepgram Nova 3 and ElevenLabs Scribe v2. The response speed competition is narrowing down to milliseconds again, and in areas where real-time response is key, like contact centers, this difference is expected to translate directly into perceived quality difference.
Source Grok Voice Think Fast 2.0 and the State of Speech-to-Speech Agents
3. OpenAI Launches Enterprise Voice and Chat Agent Platform 'Presence'
On August 11th, OpenAI introduced 'Presence', an enterprise voice and chat agent platform. It assigns tasks such as billing dispute handling, insurance claim support, and internal IT inquiries to agents, while companies directly determine accessible systems and points requiring human approval. OpenAI first applied it to its own customer service lines to handle 75% of inquiries without human intervention, reducing the agent transfer rate by 15 percentage points in just 10 days. BBVA, SoftBank, and International Airlines Group (IAG) are piloting the system. Contact center automation is moving past the pilot stage and increasingly attaching to actual support lines. However, it is worth noting that since OpenAI deploys its own engineers for deployment and post-adjustment, this is closer to a managed service than off-the-shelf software.
Source OpenAI Launches Presence for Voice and Chat Agent Deployments
4. Twilio's Self-Service Voice Usage Grows Over 50%
In its Q2 earnings release, Twilio reported that self-service voice usage surged by over 50% year-over-year. Overall voice revenue also grew by more than 20%. Notably, one AI-native client saw its revenue jump from a six-figure quarterly sum to an annual run rate of $6 million, increasing its Twilio usage by 65%.
CEO Khozema Shipchandler maintained a cautious stance, stating that it is still an early market and growth can be volatile. However, market expectations are clearly changing. As more companies apply it to actual core businesses beyond simple testing, the evaluation criteria have shifted from 'how much is automated' to 'how well conversational context is maintained and integrated with existing systems.' Ultimately, having a structure that integrates owned data to boost accuracy and seamlessly hands off to a human agent when necessary is becoming the key factor driving actual adoption.
Source Twilio Sees Voice AI Adoption Surge 50%+ Amid Growing CX Buyer Caution
5. EU AI Act High-Risk Emotion Recognition Regulation Takes Full Effect on August 2
As the high-risk AI provisions of the EU AI Act took full effect on August 2nd, customer-facing emotion recognition AI was also designated as high-risk. Accordingly, obligations have become stringent, ranging from conformity assessments to establishing human oversight, logging, and prior notification, with non-compliance risking fines of up to 3% of global revenue. For reference, emotion recognition targeting employees was already completely banned last year.
With regulations tightening like this, structures where emotion recognition features can be directly controlled outside the cloud become much more advantageous. Ultimately, whether one supports on-premise deployment alongside detailed log management is expected to be the practical battleground in this regulatory landscape.
Source EU AI Act Deadline: Customer Emotion AI Becomes High-Risk in August 2026
The gravity of the voice AI competition is shifting from model performance to how well it integrates into actual customer service fields and how effectively it navigates regulations.
If you have any questions or would like to learn more about MAGO's emotion recognition and on-premise solutions, please feel free to contact us at contact@holamago.com. We will return with fresh news next week.