screencastgen

Remote TTS

HTTP proxy TTS backend that delegates synthesis to a GPU inference server.

Source: screencastgen/providers/tts/remote_tts.py

Overview

Implements the TTSBackend protocol by proxying requests to a GPU inference server over HTTP. This is the TTS piece of the CPU/GPU split architecture.

At startup it queries GET /health to discover:

output_format
max_chunk_bytes

Those values are then used by local chunking and output handling.

Class: `RemoteTTS`

Properties

Property	Source	Description
`max_chunk_bytes`	Server `/health`	Remote backend chunk limit
`output_format`	Server `/health`	Remote backend output format

Constructor

RemoteTTS(
    server_url="http://localhost:8100",
    language="en-US",
    timeout=300,
    ref_audio_path=None,
    ref_text=None,
)

On init, the backend:

normalizes server_url
optionally reads and caches reference audio bytes
queries GET /health

Methods

Method	Description
`synthesize(text, output_path)`	Send one request to `/synthesize` and write returned audio bytes

Request Formats

Without voice cloning

Sends application/json:

text
language

With voice cloning

Sends multipart/form-data:

text
language
ref_audio
ref_text when available

The server is responsible for batching compatible requests.

BackendSpec

Field	Value
Name	`remote`
Contexts	`cli`
Capabilities	`remote`, `server_managed_reference`

Extra Args

Arg	Type	Default	Description
`--tts-server-url`	`str`	`http://localhost:8100`	GPU server URL

Dependencies

Remote TTS
├── urllib.request      (HTTP)
├── TTS Base        (BackendSpec, BackendArg)
└──▶ registered in TTS Registry
     └──▶ talks to Inference Server