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
Key Questions
What is Noiz AI and how does it compare to ElevenLabs for voice cloning?
Noiz AI provides voice cloning at 80% lower cost than ElevenLabs using just 3-second samples. It is positioned as a competitive option for indie voice SaaS wrappers alongside other tools in the highlight.
What are some of the new or updated TTS models mentioned in the Voice Surge highlight?
The highlight covers Zonos 2 with voice cloning and emotion sliders, NVIDIA's lightweight Kokoro TTS at 82M parameters, MiMo V2 with free tier and streaming, and PilotTTS with emotion control. These models target low-cost and efficient voice applications.
What announcements did ElevenLabs make at the Warsaw Summit 2026?
ElevenLabs introduced Dubbing v2 for end-to-end emotion-preserving dubbing and Eleven v4 TTS supporting whisper, sing, and mid-sentence emotion shifts. They also demonstrated a live conversational agent with enterprise partnerships.
How does Modal's infrastructure guide help with voice agent development?
Modal provides tips on reducing latency via GPU snapshotting to under 10 seconds and using warm CPU buffering plus batch sorting for cost savings. These patterns apply directly to building efficient low-cost voice SaaS wrappers.
What is Cohere Transcribe and why is it notable for ASR tasks?
Cohere Transcribe is an open-source 2B parameter ASR model that ranks #1 on the HF Far-Field ASR benchmark and supports 14 languages. It serves as a strong low-cost building block for indie voice wrappers and related 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.