Speech, Voice and Realtime Audio Models — 2026-10-25
Block has released a locally runnable voice model that sounds "eerily human," sparking both praise and fraud concerns. Mistral AI launched Voxtral 3, a full voice-agent SDK, while ElevenLabs introduced its new "Eleven Realtime" speech-to-speech model. Meanwhile, South Korean telecom giants SKT, KT, and Kakao are rolling out their government-backed "AI for All" beta services, and voice cloning scams remain a high-profile security issue.
Speech, Voice and Realtime Audio Models — 2026-10-25
Top developments
Block releases hyper-realistic local voice model
On October 24, 2026, Block (formerly Square) announced a new voice AI model capable of running locally on a MacBook M4 Max. The model has impressed users with its human-like quality but has simultaneously alarmed fraud watchers who fear it could fuel voice-cloning scams and deception. The release highlights the growing tension between accessible high-fidelity voice synthesis and the risks of deepfake fraud.
Mistral AI launches Voxtral 3 voice-agent SDK
Mistral AI released Voxtral 3 on October 20, 2026, a comprehensive suite for building multilingual voice applications. The SDK integrates speech-to-text, text-to-speech, and a voice-agent framework designed to let AI "listen, think, speak, and get things done" in real time. Transcription services are priced aggressively at $0.0035 per audio minute, positioning the tool as a cost-effective alternative for developers needing full-stack voice capabilities.
ElevenLabs debuts "Eleven Realtime" speech-to-speech model
ElevenLabs announced "Eleven Realtime," a new speech-to-speech model optimized for voice agents, on October 21, 2026. The model is available in two variants: "v1 Mini" for natural, low-latency conversations and "v1 Max" for complex reasoning tasks. It features compatibility with the OpenAI Realtime API, aiming to streamline the integration of high-quality voice into existing agent architectures.
South Korean telcos launch "AI for All" beta services
SK Telecom, KT, and Kakao have begun beta testing their government-backed "AI for All" services, with full launches targeted for October and December 2026. The services aim to provide free, universal access to AI assistants that go beyond chatbots to execute real-world tasks like reservations and payments. SKT is leveraging its "A.X K2" model to support voice and text interactions, while KT plans to integrate its "Sajang-easy" platform for small business owners.
Local view
In Japan, GIGAZINE reported on "OmniVoice," a new open-source TTS model supporting over 600 languages with zero-shot capabilities, reflecting the strong interest in accessible, multilingual voice synthesis tools. Meanwhile, South Korean media outlets like Seoul Economic Daily are closely tracking the rollout of the "AI for All" initiative, noting the strategic importance of integrating AI into everyday telecom services to compete with global giants like Google and OpenAI.
Context & numbers
The voice AI market is experiencing rapid growth, with projections estimating a $13.67 billion market size and 42% growth in voice agent adoption. ElevenLabs has reportedly reached a $22 billion valuation, underscoring the sector's high stakes. However, this growth is accompanied by significant security risks; the FBI reported $893 million in losses from AI scams in a recent period, with voice cloning scams becoming a primary vector.
On the radar
- Voice Cloning Scams: Recent reports from Fortune and ZDNET highlight the ease of cloning voices using free AI tools, with experts recommending "safe words" as a countermeasure against fake kidnapping scams.
- Deepfake Readiness Drills: Organizations are increasingly urged to conduct deepfake readiness drills to test callback and dual-approval controls against spoofed payment requests.
- Open TTS Leaderboard: A new "Open TTS Leaderboard" has launched to standardize evaluation of the 8,000+ TTS models on Hugging Face, addressing the fragmented evaluation landscape.
This content was collected, curated, and summarized entirely by AI — including how and what to gather. It may contain inaccuracies. Crew does not guarantee the accuracy of any information presented here. Always verify facts on your own before acting on them. Crew assumes no legal liability for any consequences arising from reliance on this content.
Powered by
