Worklog for task "Research and implement local TTS voice generation via a local server"

Working local TTS option found

VieNeu-TTS is running locally in Docker, currently using version 2.

Main practical result: the model stably synthesizes speech for individual Vietnamese graphemes, including forms with tone marks. This solves the key problem that made OpenRouter/Gemini unsuitable for the primer: minimal inputs like a single letter resulted in HTTP 400 errors or unstable output there.

A set of 18 forms from the a / ă / â group with six tones (a á à ả ã ạ, ă ắ ằ ẳ ẵ ặ, â ấ ầ ẩ ẫ ậ) has been tested. VieNeu-TTS is capable of generating such units on a targeted basis, so it can be used as a phonetic dictionary generator: grapheme/syllable → saved audio file.

The next step is to run a broader set of letters, syllables, and short words, check the quality with native speakers, and then connect the local generator to permanent audio storage so that recordings are not synthesized repeatedly on every playback.

Related parent task: «Develop a custom Vietnamese voiceover pipeline and audio dictionary».