Inworld AI
Low latency voice and speech infrastructure built for game characters
Inworld AI focuses on the voice layer of game characters rather than trying to be an entire game engine on its own. Its two headline products are Realtime TTS-2, text to speech you can direct with plain language instructions and tune for first audio as fast as 50 milliseconds on the Flash tier, and Realtime STT-1, streaming speech to text that also detects emotion, speaking style, age, and accent as it transcribes. Both support roughly 500 languages and dialects. On top of that sits speech to speech, a single WebSocket or WebRTC connection that listens, reasons with a language model of your choice, and speaks back, plus Inworld Inference, which gives access to open models like GLM, Gemma, MiMo, and DeepSeek through an OpenAI compatible API. It integrates with LiveKit Agents without needing a second API key, and Unity is listed among its partners, though the public page does not describe dedicated NPC dialogue tooling beyond the voice stack itself. Independent feedback is thin here. Inworld's Trustpilot page shows a 3.3 out of 5 TrustScore, but that is built from only 4 reviews in the last 12 months, too small a sample to treat as a solid signal either way. Two of those reviews praise the voice quality directly, while the negative ones describe real friction: one user reports being locked out of a paid subscription with slow support, and says they were banned from the company's Discord after complaining about it.
- Category: Games
- Pricing: Freemium
- Rating: 3.3 / 5 (4 reviews)
- Platforms: Web, API
Key features
- Realtime TTS-2 and a faster TTS-2 Flash tier, directable with plain language, first audio as fast as 50ms
- Realtime STT-1 streaming speech to text with emotion, style, age, and accent detection
- Speech to speech over one WebSocket or WebRTC connection with a choice of underlying LLM
- Inworld Inference access to open models including GLM, Gemma, MiMo, and DeepSeek via an OpenAI compatible API
- Support for roughly 500 languages and dialects
- Native LiveKit Agents integration with no second API key required
Pros & Cons
Pros
- Covers text to speech, speech to text, and speech to speech in one platform instead of stitching together several vendors
- Very low latency claims on the Flash tier, as fast as 50 milliseconds to first audio
- OpenAI compatible API formats make it easier to drop into existing tooling without a rewrite
- Broad language and dialect coverage for a single voice platform
Cons
- Trustpilot sample is tiny, only 4 reviews in the last 12 months, so the 3.3 score should not be read as a solid, settled signal
- A negative reviewer describes being locked out of a paid subscription, encountering slow support, and being banned from the company's Discord after complaining
- Pricing is entirely usage based, billed per million characters or tokens, rather than a flat predictable subscription
- The public page does not describe dedicated NPC dialogue or behavior tooling beyond the voice stack itself, despite games being one of its stated use cases
FAQs
Is Inworld AI just for games?
No, it is positioned as general voice infrastructure, with games and narrated media as one use case alongside coding assistants and other applications.
How is Inworld priced?
It is usage based, charged per million characters for text to speech and per million tokens for inference, with lower rates on higher growth and enterprise tiers rather than flat subscription pricing.
Can I trust Inworld's Trustpilot rating?
Treat it cautiously. The 3.3 out of 5 score comes from only 4 reviews in the past year, which is too small a sample to draw a firm conclusion from.