Voice Surge: OpenAI/Voxtral/Noiz/Grok/VibeVoice/Supertonic + STT Integrations + Dograh/Phonon/StepAudio/Zonos + MiMo V2 TTS + Gemini 3.1 Flash TTS + VoxCPM + Kokoro TTS + PilotTTS + Miso One + Microsoft MAI-Voice-2 + Modal Infrastructure Tips + Velma API + ElevenLabs Dubbing v2 & Eleven v4 TTS + Krisp Voice Translation API + Cohere Transcribe + Zonos 2 + MiniMax Speech Turbo 2.6 + Cartesia Sonic 3.5 & Ink 2 + Grok TTS
Key Questions
What new TTS options are available from xAI?
Grok TTS from xAI supports 5 expressive voices and over 20 languages, generating high-fidelity spoken audio with no pricing details released yet.
What are the features of StepAudio 2.5?
StepAudio 2.5 is a unified ASR/TTS/live dialogue model trending on Hugging Face with 3% CER and 68% win rate, positioning it as a strong low-cost option for voice SaaS wrappers.
How does Zonos 2 improve on previous TTS versions?
Zonos 2 adds voice cloning and emotion sliders, now available on HF Spaces as an updated release from Zyphra for expanded low-cost TTS options in indie voice applications.
Lightning V3.1 Pro TTS benchmarks; Noiz AI cloning (80% cheaper than ElevenLabs, 3-second cloning), xAI Grok TTS, VoxCPM2, VibeVoice, Supertonic 3 (31 languages, CPU-friendly). Palabra.ai signals $1M ARR demand for real-time voice translation wrappers. StepAudio 2.5 unified ASR/TTS/live dialogue model trending on HF (3% CER, 68% win rate) – strong contender for low-cost voice SaaS wrappers. Dograh open-source Vapi-style voice AI platform with Docker Compose and visual workflow builder. Phonon TTS achieves 1.00% WER with voice cloning at ~100M params, private beta. Zonos v0.1 TTS on HF Spaces with voice cloning and emotion sliders – beta release. Now updated to Zonos 2 with improved quality. MiMo V2 TTS API launched with free tier, style control, streaming, OpenAI-compatible. New: Gemini 3.1 Flash TTS now available on Cloudflare Workers AI – simple API, multiple voices, low-cost pay-as-you-go inference. VoxCPM (OpenBMB) voice cloning tool with 120-second cloning and 1M token input processing – practical for indie voice SaaS wrappers. Latest: NVIDIA released optimized Kokoro TTS (82M params) on Hugging Face – extremely lightweight, ideal for low-cost indie voice wrappers. New additions: PilotTTS (emotion control, voice cloning, 8GB VRAM, API format) and Miso One (8B TTS, 110ms latency, emotional range). Also Microsoft MAI-Voice-2 TTS API: voice cloning, emotional control, 15 languages, $22/M chars, Azure integration – strong alternative for multilingual voice agents. Modal's deep-dive on voice agent infrastructure provides actionable latency optimization (GPU snapshotting 25-90s to <10s, warm CPU buffering, canned intro tricks) and cost reduction patterns (batch sorting, isolated scaling) – directly applicable to low-cost voice SaaS wrappers. Rishabh Bhargava's voice agent engineering article provides practical latency/intelligence/reliability insights. New: Velma API from Modulate detects 150+ emotional clues from raw audio, 10x cheaper than LLM-based analysis, target $0.02/min – valuable component for voice agent wrappers. Latest: ElevenLabs Dubbing v2 (end-to-end, emotion-preserving) and Eleven v4 TTS preview (whisper/sing/mid-sentence emotion shift) announced at Warsaw Summit 2026, plus live conversational agent demo. Telco/airline partnerships signal enterprise traction. Vobiz.ai building voice AI backbone in India with 80ms P95 latency and 98% retention – reinforces infrastructure importance for voice wrappers. Also noted: Krisp Voice Translation API (96% accuracy, enterprise-grade) – potential building block for high-accuracy voice translation wrappers. New: Cohere Transcribe open-source 2B ASR model, #1 on HF Far-Field ASR benchmark, supports 14 languages – strong low-cost building block for indie voice wrappers. New: Zonos 2 (update to Zyphra's TTS) now on HF Spaces with voice cloning – further expands low-cost TTS options for indie wrappers. New: MiniMax Speech Turbo 2.6 TTS pricing and specs now available – speed-optimized TTS, relevant for voice surge wrappers. Need to compare latency/cost against ElevenLabs, Zonos, etc. New: Cartesia Sonic 3.5 and Ink 2 are #1 streaming TTS and STT models, respectively, offering top performance for voice agent wrappers – pricing not yet disclosed. New: Grok TTS from xAI with 5 voices, 20+ languages, no pricing yet – worth monitoring for cost and quality.