{% extends "base.html" %} {% set labels = { "welcome": "Welcome", "media": "Media folders", "vocabulary": "Vocabulary", "llm": "LLM enhancement", "services": "Model services", "finish": "Finish", } %} {% block title %}Voxint setup — {{ labels[step.value] }}{% endblock %} {# The wizard runs before onboarding, so the protected top nav (Runs / Review) would only bounce the operator back here. Replace it with a step indicator. #} {% block nav %}
Voxint transcribes your audio and video, separates who spoke when, and lets you attribute each voice to a real person — so you get a searchable, speaker-labelled transcript you can trust.
This quick setup takes about two minutes. Every step is optional and you can revisit it later.
{% elif step.value == "media" %}Voxint reads media from under {{ media_root }}. Register one folder per
line (paths relative to that root); leave blank to skip.
Registered: {% for f in media_folders %}{{ f }}{% if not loop.last %}, {% endif %}{% endfor %}
Optionally look through your registered folders for audio/video that isn't in Voxint yet, and queue it for transcription.
{% elif step.value == "vocabulary" %}Add names, jargon, acronyms, and preferred spellings — one per line. Voxint feeds these to transcription and to name attribution, so unusual terms come out right. They apply to your next run with no restart needed.
{% elif step.value == "llm" %}When enabled, Voxint sends transcript segments to an OpenAI-compatible model to clean them up and surface likely speaker names. It's best-effort — a slow or failing model never blocks a run; enhancement is simply skipped.
{% if llm_key_present %}An LLM_API_KEY is set in the environment. ✓
No LLM_API_KEY is set. The key is configured in the
environment (never stored here); set it and restart the worker before enabling.
The configured LLM run budget doesn't fit the transcription stage lease, so enhancement can't be enabled until that's adjusted.
{% endif %} {% elif step.value == "services" %}Voxint's transcription, diarization, and speaker-embedding run as separate model services (GPU or CPU tier). Here's whether each is reachable right now:
{% for svc in services %}If a service is down, start the model services — the GPU overlay
(compose.gpu.yaml) or, with no NVIDIA GPU, the CPU overlay
(compose.cpu.yaml); see the README's quickstart. This check is advisory —
you can finish setup regardless. But a run submitted while a service it needs is down
will retry with backoff (about five attempts over roughly an hour and a half) and then
land failed; once the services are up, requeue it from the run's
page.
Finishing starts a short guided tutorial on a bundled three-speaker sample. You can re-run this setup — or the tutorial — any time from Settings.
{% else %}Finishing opens the review console. You can re-run this setup any time from Settings.
Tip: run voxint tutorial seed to enable the guided
tutorial on a bundled three-speaker sample.