Gandr
Gandr provides an unlimited, low-latency text-to-speech and instant voice cloning API for developers and enterprises, with streaming support via WebSocket or HTTP. The platform clones a voice from a roughly ten-second reference clip without a training job or per-voice fee ante, and delivers first audio in 146 ms over the open internet. Every generated clip carries an inaudible watermark, and customer audio is never used for training.
- Artificial Intelligence
- AI Agents
- Software Only
Funding
Founders
Product
Problem
Real-time voice applications—such as AI agents, live translation, and interactive assistants—need text-to-speech that is fast enough for natural conversation and inexpensive enough to scale. Traditional TTS providers charge per character, meter usage, and require lengthy training jobs or per-voice fees for cloning, making them impractical for high-volume, streaming use cases. Latency tail behavior is often poor, degrading conversational quality precisely when it matters most.
Solution
Gandr offers Spex-TTS, a production-grade text-to-speech API built around flat-rate pricing and millisecond-level latency. The platform supports streaming via WebSockets and server-sent events, plus one-shot WAV output, with a drop-in request shape compatible with existing systems. Voice cloning is instant: a ten-second reference clip becomes a working voice handled as a request parameter, with no training pipeline or per-voice charge, and the cloned identity is preserved across all 23 supported languages. The service runs on a redundant three-tier architecture behind a single endpoint, achieving 116 ms p50 server-side time to first audio. Every generated clip is stamped with an inaudible watermark, and customer data never enters a training set. Gandr also provides a live console on its homepage that demonstrates the production API in real time.
Target Audience
Primary customers are developers, startups, agencies, game studios, and enterprises building AI voice agents, customer support systems, live translation tools, video dubbing pipelines, and audiobook narration workflows that require real-time or high-volume synthetic speech with low latency.
Features
- Instant voice cloning from a 5–10 second reference clip, with fingerprinting and caching so the first request pays the cloning work and subsequent uses skip it
- Streaming audio via WebSocket for conversations, server-sent events for long reads, and one-shot WAV for batch requests, all under a single endpoint
- 23-language support with cross-lingual identity preservation, so one clone sounds like the same person in every language
- Low-latency serving: 116 ms p50 first audio server-side, 146 ms over the open internet, with published benchmarking methodology covering the full request path
- Inaudible watermark applied at generation on every request, marking audio as synthetic and surviving re-encoding
- No training on customer data: reference audio and text never enter a training set, with per-key usage auditable via API
- Drop-in REST API with a simple request shape, compatible with an x-api-key header and returning standard audio formats like 24 kHz WAV