tts-service

Hungarian text-to-speech in a multi-tenant FastAPI app is provided by exposing a /speak endpoint with ONNX-backed synthesis. It loads Piper voice models to deliver per-tenant speech synthesis from a single shared deployment.

Piper ONNX models are used for real-time voice synthesis by loading PiperVoice from configured ONNX and JSON file paths. The FastAPI service processes text requests and returns base64 audio or binary WAV responses.

Configuring a Piper voice FastAPI service requires the PIPER_VOICE, PIPER_VOICE_JSON, and TTS_DEVICE environment variables. These parameters load the ONNX model and JSON config to ensure deterministic startup and per-tenant voice consistency.

A shared TTS service for multiple tenants with consistent voices is deployed via a FastAPI server with per-tenant voice configuration. It uses Piper ONNX models controlled by environment variables to maintain deterministic synthesis across tenants.

The Hungarian TTS service returns both base64 audio payloads and binary WAV files through its /speak endpoint. This allows flexible integration for chatbots delivering spoken responses through web browsers or mobile apps.

ONNX-backed voice synthesis for multi-tenant platforms requires the Piper library and correctly configured voice model paths. It is specifically designed for Hungarian language synthesis and depends on environment variables for deterministic runtime behavior.