insonic v1.1.0

insonic accepts audio and video inputs, retains canonical audio, correlates subtitles, tracks speakers and explores what was said and when through a CLI and desktop wrapper. Replaceable local and hosted tools operate through explicit adapters while the user controls the library, configuration and processing history. Source-linked speaker audio can optionally feed trained models associated with those speakers.

v1.1.0 documents this release. Package availability and qualification are recorded in its release manifest.

Read in this order

1. Product and requirements states the outcome and required behavior.
2. Architecture explains authority, processing and recovery boundaries.
3. Technology decisions compares implementation choices and dependency qualification.
4. Delivery outcomes explains capability dependencies.

Optional first-run setup covers editable library location, backend choice before provisioning, saved progress, processing/services and first import. Skip or defer any optional stage, then resume through either client.

Local processing setup covers managed CPU environments, verified model reuse, offline acquisition and task-specific readiness.

CLI invocation and results covers contextual help, argument parsing, JSON files, output streams and durable failure inspection.

Focused contracts cover import dates and metadata, media, pipelines, subtitles, speakers, speaker audio and models, queries, storage, catalog structure, JSON contracts, credentials, credential commands and downloaded models, desktop behavior and development. The glossary defines technical vocabulary with learning and local usage links. The changelog records changes.

The guiding outcome

A user adds media with original metadata and optional recording dates/subtitles, selects an understandable preset and receives a searchable library. Results link to the original media position and distinguish acoustic voices, named speakers, transcript wording and extracted assertions. The CLI exposes core operations and machine-readable results; the GUI wraps the shared runtime and provides a media timeline and saved graph views. Optional speaker-model training uses a corpus of valid current evidence references without interrupting ordinary library work.

flowchart TB
  Import[Audio or video with optional subtitles] --> Library[Canonical audio and captured source facts]
  Library --> Pipeline[User-selected local or hosted pipeline]
  Pipeline --> Cues[Correlated Cueson transcript and speaker evidence]
  Cues --> Query[Search and source-linked graph queries]
  Cues --> Corpus[Speaker-linked audio corpus]
  Corpus --> Training[Optional elected model training]
  Training --> Models[Versioned speaker models and CLI retrieval]
  Query --> CLI[CLI results]
  Query --> GUI[Desktop timeline and saved graph views]