Inworld AI

Overview

Inworld combines its own text-to-speech and speech-to-text models with customer-selected language models. Its Realtime API handles speech input and output over a single WebSocket, while Realtime Router can direct requests among OpenAI, Anthropic, Google, and more than 220 models. That breadth makes Inworld more than a voice generator: it is a building block for applications that need live speech and model choice in one stack.

The trade-off is that the offer is oriented toward developers and product teams rather than people seeking a standalone desktop voice editor. OpenAI compatibility and a Twilio integration for Realtime TTS extend its reach, but game-engine support is at different stages: Unreal AI Runtime is available to download, while the Unity SDK is in early access.

Key features

Realtime speech

Realtime TTS-2 supports voice direction, voice cloning, and 200+ languages; Inworld states time to first byte is under 100 ms. Realtime STT streams transcription and can provide timestamps, diarization, and voice-profile signals such as emotion, age, accent, pitch, and style, alongside custom vocabulary. Those controls suit conversational and media applications that need more than a basic text-to-audio endpoint. The stated speed is useful context, but it is not a guarantee of end-to-end responsiveness in a particular application.

The combined Realtime API carries speech in and speech out over one WebSocket and supports custom voices and tool calling. For teams connecting speech to application actions, that consolidates key pieces of a voice interaction; teams needing only occasional audio generation may not benefit from the broader API surface.

Model routing and security

Realtime Router spans OpenAI, Anthropic, Google, and 220+ models. Inworld says it adds zero markup and includes failover, A/B testing, and analytics, making it a practical option for teams that want routing choices without limiting themselves to one provider. The security offering includes SOC 2 Type II certification, HIPAA and GDPR compliance, and AES encryption at rest and in transit. Customers can ask Inworld to enable zero data retention; Growth lists ZDR as an add-on rather than an included default.

Pricing

On-Demand is free to start, with usage-based rates, up to 70 minutes of TTS included, 100 custom voices, and community support. After the included allowance, TTS-2 costs $25/1M chars, TTS-2 Flash $15/1M chars, and STT 1 $0.15/hr; LLMs are billed at cost. This is the lowest-commitment way to try the service, but ongoing use is not a flat-rate subscription.

Paid plans begin with Creator at $25/mo. It supplies $25 in monthly credits, up to 33% off TTS and STT rates, 500 custom voices, and 40K characters per TTS Playground request. Unused subscription credits roll over for up to three months while the customer maintains an active paid plan of equal or higher value. Creator suits smaller projects that can use its monthly credit balance; the credit is not an unlimited usage allowance.

Builder costs $100/mo and includes $100 in credits, up to 40% off TTS and STT rates, 3,000 custom voices, increased concurrency limits, and community support. Developer costs $300/mo, with $300 in credits, discounts up to 47%, 10,000 custom voices, and priority email support. Growth costs $1,500/mo and raises those figures to $1,500 in credits, discounts up to 53%, and 30,000 custom voices, with higher API concurrency and limits. It adds priority email support; ZDR, HIPAA, and BAA are add-ons. These tiers make progressively larger workloads and voice libraries easier to accommodate, but their monthly commitments only make sense when usage and capacity justify them.

Enterprise has custom pricing, limits, and terms. It offers TTS-2 as low as $5/1M, TTS-2 Flash below $5/1M, custom STT and LLM pricing, SLA and DPA, EU and India data residency, and a dedicated account manager and Slack channel. It is the tier to discuss when contract terms, residency, or dedicated support matter; teams should compare its negotiated terms with the published tiers.

Platforms

Inworld supports API, web, and self-hosted use. It offers audio export in MP3, WAV, PCM, LINEAR16, OGG_OPUS, μ-law, A-law, and FLAC. Unreal developers can download its AI Runtime SDK; Unity developers should account for the SDK’s early-access status before planning around it.

Who it's for

Inworld targets consumer-facing applications, with solution areas spanning social and companion products, learning and education, health and wellness, agentic workforce, and games and media. Its breadth is most relevant to teams building interactive products that combine speech with model routing. It is less compelling for a casual user who wants a simple voice tool without an API-oriented workflow.

Pros and cons

• Pros: TTS, streaming STT, and a speech-in/speech-out API cover several parts of a realtime voice experience in one service.
• Pros: Router access across 220+ models, with failover and A/B testing, gives teams flexibility beyond a single LLM provider.
• Pros: The free On-Demand entry includes up to 70 minutes of TTS and 100 custom voices, so evaluation need not start with a monthly plan.
• Cons: On-Demand turns into usage-based billing, while meaningful discounts and higher concurrency require monthly commitments up to $1,500/mo.
• Cons: Unity support is early access, so it is a less settled fit for teams that need a production-ready Unity SDK today.
• Cons: ZDR and HIPAA/BAA on Growth are add-ons, not included plan entitlements, which adds a qualification step for teams with those requirements.

Alternatives

For game-specific discovery, browse Game Voice Generators. Choose Voice Creator Pro if you prefer a freemium option spanning macOS, web, and Windows; its free plan includes 5,000 tokens monthly, about 1.5 hours of audio, and 30-day file retention. ElevenLabs is another API-and-web freemium choice, with a free plan offering 10,000 credits per month and three Studio projects.

Altered Studio may fit better when desktop platforms and voice morphing are priorities; its free plan includes three minutes per month of Voice Morphing, 10,000 AI Tokens, local voice cloning, and a noncommercial attribution license. Choose Google Cloud Text-to-Speech if you want a paid API-and-web service with free usage and per-character pricing; its Standard voices include 0 to 4 million characters free and cost US$4 per 1 million characters thereafter.

Neocortex is a freemium alternative available on web, Windows, macOS, Linux, iOS, and Android. PlayHT offers API and web access, with a free Starter plan that includes 10,000 credits per month, one voice-clone slot, one team seat, MP3 output, and community support. VoiceKeep is a freemium web alternative. Consider Amazon Polly for a paid API-and-web service with free access and monthly character-based Standard and Neural plans.

Verdict

Inworld AI is a strong fit for development teams building consumer-facing realtime voice products that need both speech services and flexible LLM routing. Its free On-Demand entry and broad speech feature set are the clearest reasons to choose it; look elsewhere if you need a finished desktop voice editor or a fully available Unity SDK, or if the monthly commitments required for higher capacity do not match your workload.