AI AGENT STORE / SERVICES

MEDIA PROCESSING & DELIVERY · FOR AI AGENTS

Inworld TTS

Generate speech from text using hosted voice models and streaming audio APIs.

Inworld TTS turns text into audio for spoken agent responses, narration and accessible interfaces. REST and Python or JavaScript SDKs support streaming and non-streaming synthesis with selectable voices and audio formats.

Your integration checklist

What you need from your owner
Create an account, choose a voice and usage plan, and issue a Standard API key. Obtain the necessary permission before using a custom voice.
Authentication
Server requests use Authorization: Basic with the complete Base64 credential copied from the portal as INWORLD_API_KEY; do not encode it twice. Client applications should obtain short-lived tokens through a backend.
Your first task
Synthesize a short sentence containing a name and a number, then check the audio format and pronunciation in the intended playback client.
Billing & payment
Usage consumes dollar-denominated credits. TTS rates depend on model and plan; paid plans include credits and overages use the plan rate. Voice cloning and other Inworld products can have separate requirements.
Agent payment support
Not verified. Programmatic usage does not establish autonomous checkout support.
Operational limits
Streaming chunks are format-dependent. WAV streams have one initial header, while LINEAR16 chunks contain individual WAV headers. Pronunciation, latency and language suitability need evaluation on the intended content.

ORIGINAL AI AGENT STORE INTEGRATION PLAYBOOK

Validate the audio stream before connecting it to a conversation

Editorial recommendation · Reviewed 2026-11-08

When to choose this service

Choose hosted synthesis when an agent needs repeatable voice output without operating speech models. Existing device speech output can suffice for a simple local assistant.

Maintained speech models, voice assets and serving infrastructure provide audio generation capacity beyond text reasoning. A client script alone does not supply the trained model or hosted inference.

Suggested workflow

  1. Choose a voice, model and output format for the playback device; start with a short representative sentence.
  2. Test names, numbers and punctuation. Check chunk boundaries and sample rate before adding live text generation.
  3. Measure response time in the actual application and track characters and credits per task. Save the source text with approved audio assets.

Acceptance checks before real work

  • A valid audio file can still contain an incorrect spoken number.
  • A text stream boundary is not necessarily a useful speech boundary.
  • Appending chunks requires the correct handling for their audio container.

If the workflow fails

For clicks or corrupted playback, inspect the encoding and header handling before repeating synthesis. For pronunciation problems, revise the spoken form and test a short sample.

Provider documentation

Explore services for adjacent needs

Continue building your toolkit