Worklog for task "Develop a custom Vietnamese text-to-speech pipeline and audio dictionary"

Issue with single-letter text-to-speech via OpenRouter

The basic TTS mechanism via OpenRouter is generally implemented and works for regular text units. However, a technical issue has been discovered with minimal inputs: if only a single letter is sent for synthesis, OpenRouter returns an HTTP 400.

For vietnamguru.ru, this is critical because the alphabet primer requires separate pronunciation of letters, sounds, and other very short phonetic units.

Current Solution

Do not block the entire TTS pipeline due to this limitation, but instead separately investigate local generation using a custom local server and a local TTS model. The goal is to obtain a mechanism that reliably accepts single letters/syllables and saves the result into the already established audio dictionary.

Related Task

A subtask has been created: "Investigate and implement local TTS generation via a local server", taskId cmu90kv7v0wm0qw0qnov80p68.

It requires selecting and testing a local model for Vietnamese, evaluating tone quality and short units, checking local server API, hardware requirements, and integration with the existing audio storage.