nodecode-voicestudio
VoiceStudio: speak, design voices, render chaptered audiobooks and dub videos through a local VoiceStudio server. A Nodecode rewrite of aska-digital/protean-voice-studio by aska-digital.
Six tools, and a command for each (and /voicestudio, the cell's own): vs_speak says a text and names the audio file, vs_list_voices lists the voices the server offers, vs_design_voice makes a new one from words or from a recording, vs_longform renders a chaptered script into an audiobook, vs_dub dubs a video into other languages, and vs_status says whether the server is ready.
The speech is made by the VoiceStudio app on this machine. The cell only talks to the server it runs, at the address the section names; nothing else is contacted, and the audio it answers is written under the cell's own folder.
Get it
Tell the agent "add the voicestudio cell", or run:
nodecode add nc://voicestudio
Set it up
Install and start VoiceStudio and let it finish loading its models. Its server answers on http://127.0.0.1:3900 unless the app says otherwise; put that address in the section.
The section, as ~/.nodecode/config.jsonc carries it:
| Setting | Default | What it does |
|---|---|---|
base_url |
http://127.0.0.1:3900 |
the address the VoiceStudio app answers on |
default_voice |
(unset) | the voice a call names none; /vs-voices lists the ids |
directory |
voice-studio/ |
where the audio files are written, under the home |
timeout |
600 |
seconds one generation may take |
connect |
3 |
seconds a call waits for the server to answer at all |
probe |
3 |
seconds one status or job read waits |
poll_interval |
10 |
seconds between two job reads |
polls |
60 |
how many times a job is read before the cell answers that it is still going |
Without a server the cell still starts: every tool then answers that it could not reach VoiceStudio and names the address. vs_status says when it is back.
Use it
Ask for it: "say this out loud in the narrator voice", "design a warm young female voice called Ana", "which voices have I got", "render this script as an audiobook", "dub this clip into Spanish and French", "is VoiceStudio ready".
| Tool | Arguments | What a user gets |
|---|---|---|
vs_speak |
text, voice, speed, seed, out_path |
the audio file, its size, the voice and the seed that repeats the take |
vs_list_voices |
search, kind |
the saved voices, the gallery and the presets, with the ids the other tools take |
vs_design_voice |
name, describe or sample, preview |
the new voice's id, and a rendered preview unless preview is false |
vs_longform |
file, voice, format, title, author, resume, seed |
the chaptered file, its format and its chapter count |
vs_dub |
video, langs, voice |
the file each language made, and which language failed when one did |
vs_status |
detail |
whether the server is ready, its engine, its voice count, its free disk |
Commands take the same arguments as --flag value, and quote their text. /voicestudio prints the commands beside the settings and the address the cell runs on, and --status asks the server itself:
/voicestudio /voicestudio --status /vs-speak --text "Ship it." --voice narrator /vs-design --name Ana --describe "warm young female, slight British accent" /vs-voices --search warm --kind mine /vs-longform --file episode.md --format m4b --title "Episode one" /vs-dub --video intro.mp4 --langs es,fr /vs-status
A script for vs_longform is markdown: each # heading starts a chapter and ## stays body text. The markup a voice can perform is small -- [pause], [pause 500ms], [slow], [fast], [emphasis], [spell], [[word|say-this]] and [voice:NAME]. Anything else ([sigh], <break/>, (pause)) is refused before the render, with the lines named, because a voice would read it aloud literally.
A render or a dub the server has not finished is reported with its job id: run the tool again with that id in resume and nothing is lost.
