Latest developments in TTS and voice AI.
Gemini 3.1 Flash TTS introduces 200-plus inline audio tags, scene direction, speaker-level audio profiles, native multi-speaker dialogue, and 30 prebuilt voices across 80-plus languages, available today in preview through the Gemini API, Vertex AI, and Google Vids.
Smallest.ai released Lightning V3 on March 27, 2026, a non-autoregressive TTS model claiming a 3.89 MOS and 76% preference over OpenAI. The benchmarks are self-reported, the model is closed-source, and the press coverage is entirely wire syndication.
Mistral AI releases Voxtral TTS with 9-language support, 3-second voice cloning, and open weights on Hugging Face. Priced at $0.016 per 1,000 characters.
Nine months after enforcement began, the EAA is driving measurable growth in TTS API usage across European markets.
IBM adds Deepgram's speech-to-text and text-to-speech to watsonx Orchestrate for multilingual enterprise voice.
Alibaba's Qwen team released Qwen3-TTS under Apache 2.0 with 97ms latency and 10-language support.