Database

Sirene uses PocketBase as its database, file storage, and real-time engine. It runs inside the sirene binary; the dashboard is served at /_/ (port 8090 in development) and requires a superuser created with sirene superuser upsert.

Collections

Collections are created automatically via the Go migrations in server/migrations/.

voices

Field	Type	Description
name	text	Voice name
description	text	Description
language	text	Language code (en, fr, es...)
avatar	file	Voice avatar image
builtIn	bool	Built-in model voice vs custom
samples	relation[]	Link to voice_samples (multi-select, cascade delete)

voice_samples

Field	Type	Description
audio	file	Audio file (WAV/MP3)
transcript	text	Sample transcription (manual or auto via Whisper)
duration	number	Duration in seconds

generations

Field	Type	Description
voice	relation	Link to voices
model	text	Model ID (e.g., kokoro-v1)
text	text	Source text
language	text	Generation language
audio	file	Generated audio file
duration	number	Duration in seconds
speed	number	Generation speed

Voice Creation Workflow

1. User clicks "Create Voice"
   → Dialog: name, description, language, avatar

2. Upload 1-N audio samples (WAV/MP3, 5-30s each)
   → Waveform preview (wavesurfer.js), play/pause, duration
   → Transcription: text field + auto-transcribe button (Whisper)
   → Upload to voice_samples collection

3. Sirene server:
   a. Creates the voice in PocketBase (voices collection)
   b. Stores samples in PocketBase (voice_samples, file field)
   c. Forwards samples to the Python service for preprocessing

4. Python service:
   a. Decodes and normalizes samples (resample, mono, RMS)
   b. Prepares optimized reference audio for each compatible backend
   c. Stores processed files in the volume

5. The voice is available for generation
   → Client sees the voice appear in real-time via PocketBase SSE

Real-time Updates

PocketBase provides real-time subscriptions via Server-Sent Events. The client subscribes to collection changes to get instant updates when:

• A new generation is created
• A voice is added or modified
• Download progress updates (via the server SSE endpoint, not PocketBase)