Reporting on voice automation, contact center systems, latency, QA, and adoption in production workflows.
Kandinsky Lab has published Kandinsky 6.0 Video Pro (29B) and Lite (3B), MIT-licensed diffusion models that generate five-second clips with synchronized 44 kHz audio, dialogue and lip-sync from text or a first frame.
Cactus Compute released Whistle on October 2, 2026. The Apache-2.0 speech-to-text model fits in one 16.9 MB file, runs on CPUs alongside its Needle tool-calling model, and transcribes seven languages with word timestamps on phones, wearables, and microcontrollers.
Alibaba’s Qwen team shipped Qwen-Audio-3.1-Realtime as a full-duplex voice stack for tool-using agents—Think, Act, and Speak and Coordinate—live on QwenCloud as qwen-audio-3.1-realtime-plus with a technical report on arXiv.
On September 24, 2026, Google introduced Gemini 3.8 Live with Live Avatar in Gemini Enterprise—near real-time speech-to-speech dialogue paired with lip-synced streaming video, multilingual across 97 languages and watermarked with SynthID.
Tencent’s Hunyuan Speech Team and academic partners released Gander, a 9B Apache 2.0 omni-interaction model for streaming speech, vision, and async agent tasks, with weights on Hugging Face and an open runtime on GitHub.
Conversational software provider Gupshup launched its self-serve Voice AI Platform to support automated inbound and outbound phone workflows. The platform extends WhatsApp, RCS, and SMS workflows into voice while allowing businesses to select custom speech-to-text, text-to-speech, and language models. The service is available globally at a starting rate of $0.035 per minute with 100 promotional minutes included.
Amazon Web Services released native configuration support allowing Amazon Lex V2 bots to use Deepgram for automated speech recognition. The system securely routes bot audio streams through Deepgram API endpoints via credentials stored in AWS Secrets Manager. The integration is available globally across AWS regions with support for custom Deepgram model IDs and multi-language acoustic tuning.
Hume AI released a new benchmark called Real World VoiceEQ to evaluate over 40 leading proprietary and open-source voice models on emotional intelligence and human-like qualities. The benchmark tests models across 15 key evaluation dimensions and 60 metrics, including speech recognition, text-to-speech quality, speech-to-speech performance, and speech understanding. The evaluations are grounded in over one million individual human ratings collected from 10,000 global participants using Hume AI's Kairos platform.
Microsoft launched its new in-house text-to-speech model, MAI-Voice-2, in a free public sandbox called the MAI Playground. The multilingual model supports 15 languages and can clone a voice from a short reference audio sample of 5 to 60 seconds. To prevent abuse, Microsoft enforces system-level consent requiring users to upload consent audio and apply for Azure review before deploying custom voices.
Developer Jamie Pine's open-source repository Voicebox spiked in popularity on GitHub, crossing 35,000 stars. Written in TypeScript and built with Tauri and Rust, the local-first application combines zero-shot voice cloning across seven text-to-speech engines with system-wide dictation. The software runs fully offline on macOS, Windows, and Linux, allowing users to give Model Context Protocol (MCP)-aware AI agents custom cloned voices under an open-source license.
Elon Musk's xAI launched Grok Voice TTS 1.0, a new text-to-speech model supporting automatic language detection. The model converts text into spoken audio across more than 20 languages and offers five built-in voices with inline speech tags to control pauses, pitch, and speed. Output is available in formats including MP3, WAV, and PCM at sample rates up to 48 kHz, with a limit of 15,000 characters per request.
