Glossary
This glossary defines the technical vocabulary used in the system contracts. Each entry links to a primary learning resource and the local pages where the term is used most heavily. External references explain the underlying technology; the linked insonic contracts define its application here.
| Term | Meaning in insonic | Learn and local usage |
|---|---|---|
| Adapter | A replaceable implementation of a declared capability, with versioned configuration, results and errors. | Go interfaces; Pipelines, architecture |
| Alias | A revisioned name for a catalog speaker, with scope and provenance; identical text does not automatically merge identities. | W3C provenance model; Speakers, schema |
| Alignment | Matching selected transcript text to audio to produce timed observations without silently replacing the text. | PyTorch alignment tutorial; Pipelines, subtitles |
| API (application programming interface) | The requests, results and errors through which components invoke shared application behavior. | OpenAPI concepts; Architecture, desktop |
| ArcadeDB | The supported server graph backend selected through its configured endpoint and database. | ArcadeDB HTTP API; Graph, technology |
| Artifact | An immutable byte object, such as original media, a metadata report, a subtitle document or model weights, with digest and location records. | W3C provenance model; Storage, schema |
| ASR (automatic speech recognition) | A model operation that turns spoken audio into text; useful correlated results also require timing. | faster-whisper; Pipelines, technology |
| Assertion | A structured proposition attributed to cited source evidence, including its polarity and conditions; storage does not establish its truth. | W3C statement model; Graph, schema |
| Attribution | A revisioned observation linking a run-local voice or source interval to a speaker, with basis, confidence and provenance. | W3C provenance model; Speakers, voice models |
| Authenticated encryption | Encryption that also detects ciphertext modification, used by an encrypted credential vault through maintained libraries. | Authenticated encryption standard; security |
| Catalog | The SQLite or PostgreSQL authority for operational state, associations, selections and user edits. | SQLite transactions; Architecture, schema |
| Cgo | Go's mechanism for calling C code, introducing native-library and platform packaging requirements. | Cgo documentation; Technology, architecture |
| Checkpoint (graph) | The last atomically committed projection event, generation and catalog revision used to reconcile publication and replay. | Transactional outbox pattern; Graph, architecture |
| Checkpoint (model) | Persisted training state or weights that permit a compatible adapter to resume or recover a training attempt. | PyTorch model persistence; Voice models, schema |
| Chunk | A bounded window of selected transcript cues prepared for an adapter, retaining exact cue and source interval references. | faster-whisper; Graph, pipelines |
| CLI (command-line interface) | The text and automation interface exposing application operations, human output and versioned JSON results. | Go documentation; Desktop, architecture |
| Codec | An encoding/decoding method for media streams whose exact implementation and licensing matter for packaging. | ffprobe documentation; Media, technology |
| Content-addressed storage | Managed object naming derived from a content digest, with separate logical IDs and physical locations. | NIST Secure Hash Standard; Storage, media |
| Corpus | Audio segments selected across the library by a speaker-linked policy; training freezes a particular membership snapshot. | OpenLineage model; Voice models, speakers |
| Correlation | The recorded link between an artifact or observation and original media, including a mapping between their time coordinates. | W3C provenance model; Media, subtitles |
| CPU / GPU | A general-purpose processor and a graphics/compute accelerator; scheduling and model profiles account for their capacity separately. | CTranslate2 hardware support; technology, pipelines |
| CTranslate2 | The inference engine used by faster-whisper, with qualified hardware and numeric-format support. | CTranslate2 documentation; technology |
| Cue | A timed subtitle unit carrying text and format-specific fields; cue boundaries need not equal acoustic speaker boundaries. | WebVTT specification; Subtitles, graph |
| Cue JSON | Cueson's schema-governed subtitle document, including source bytes, common cues and native-format fields. | Cueson schema; Subtitles, contracts |
| Cueson | The required upstream executable and schema that govern subtitle admission, normalization, validation and rendering. | Cueson schema; Subtitles, technology |
| Dataset snapshot | Immutable selected corpus membership and exact source, segment, attribution and transcript revisions for a training job. | OpenLineage model; Voice models, schema |
| Derivative | An artifact produced from a parent by recorded transforms, preserving parent identity and time/channel mappings. | W3C provenance model; Media, pipelines |
| Diarization | Acoustic segmentation of which voice spoke when, including overlapping speech; it does not itself name a person. | pyannote audio; Speakers, technology |
| Digest / SHA-256 | A cryptographic hash of exact bytes used for content identity and integrity checks; it is separate from an object-store ETag. | NIST Secure Hash Standard; Storage, media |
| DST (daylight saving time) | A time-zone rule that can repeat or skip local wall times; origination input preserves the chosen interpretation. | IANA time-zone database; Ingestion, desktop |
| Embedding | A model-produced numeric vector for an explicitly scoped audio/text input, stored with model and source provenance. | pyannote audio; Speakers, pipelines |
| ETag | An object-store transport/version validator whose meaning depends on the provider and upload; it is not an insonic SHA-256 digest. | S3 object metadata; Storage, architecture |
| Evidence span | The exact cue references and original-media intervals supporting a result or assertion. | W3C provenance model; Graph, schema |
| EXIF (Exchangeable Image File Format) | An embedded metadata family whose available tags are retained alongside container, stream and other observations. | ExifTool tag reference; Ingestion, media |
| faster-whisper | A Whisper transcription implementation using Python and CTranslate2, run through a managed or connected worker. | faster-whisper project; technology |
| Fencing generation | A monotonically increasing ownership token checked at the write destination so an older worker cannot commit after takeover. | Distributed locking and fencing; Architecture, graph |
| FFI (foreign function interface) | A boundary for calling code built in another language, requiring compatible types, memory ownership and native packaging. | Rust FFI guide; technology |
| FFmpeg / ffprobe | Media tools for stream inspection, decoding and transforms, packaged with exact versions and applicable notices. | ffprobe documentation; Technology, media |
| Force-directed graph | A visual layout using simulated forces to position selected nodes and edges; layout positions are separate from graph facts. | D3 force simulation; Graph, desktop |
| Foreign key | A database constraint relating one record to another; workspace-scoped relationships also prevent cross-workspace references. | SQLite foreign keys; Schema, storage |
| Go | The orchestration language for the shared application/runtime and CLI; model workers can use other runtimes. | Go documentation; Technology, architecture |
| Graph projection | A rebuildable LadybugDB or ArcadeDB representation of accepted evidence and catalog selections. | LadybugDB documentation; Graph, architecture |
| GUI (graphical user interface) | The desktop wrapper for shared runtime operations, including playback, timeline and saved graph views. | Wails introduction; Desktop, technology |
| Half-open interval | A source range including its start and excluding its end, written [start, end), so adjacent boundaries do not overlap by definition. | WebVTT specification; Schema, media |
| Idempotency | The ability to repeat an exact operation without duplicating its effects, using recorded operation IDs and receipts. | HTTP idempotency; Architecture, storage |
| Immutable | Recorded bytes or evidence that are not overwritten; changes create a new version or selection revision. | W3C provenance model; Schema, media |
| Inference | Running a selected model on input to obtain predictions; it differs from training new model weights. | faster-whisper; Pipelines, voice models |
| IPC (interprocess communication) | The local connection between clients and the owning runtime, using a user-only socket or named pipe and version handshake. | Unix sockets; Architecture, security |
| JSON (JavaScript Object Notation) | A structured text interchange format used for CLI results, configuration, manifests and contract records. | JSON standard; Contracts, desktop |
| JSON Schema | A machine-readable description of JSON structure and validation constraints; cross-record semantic integrity still needs runtime checks. | JSON Schema guide; Contracts, schema |
| KDF (key derivation function) | A function deriving cryptographic key material from a password and parameters; a vault records its compatible algorithm settings. | NIST password-based key derivation; security |
| LadybugDB | The default embedded property-graph and openCypher engine, owned by one writable workspace runtime. | LadybugDB documentation; Graph, architecture |
| Lease | A time-bounded ownership claim for jobs, materializations or publication; expiry alone does not fence an old destination writer. | Distributed locking and fencing; Storage, schema |
| Lineage | Recorded relationships from source through transforms, datasets and model versions, preserving origins after identity edits. | OpenLineage model; Voice models, schema |
| LLM (large language model) | A text-generating model used through declared reasoning/query capabilities; a chat endpoint does not automatically provide acoustic recognition. | Hugging Face text generation; pipelines, graph |
| Logical schema | Backend-neutral entities, IDs, revisions and integrity rules implemented by backend-specific tables, indexes and migrations. | JSON Schema guide; Schema, storage |
| Manifest | A structured inventory of inputs or artifacts and their identities, options, digests or inclusion status. | OpenLineage model; Ingestion, storage |
| Materialization | A verified local copy of an artifact made available to a tool, with a lease and separate cache location. | S3 object storage; Storage, pipelines |
| Media asset | A particular byte sequence and stream selection, independent of the library entries that reference it. | ffprobe documentation; Media, schema |
| Media entry | A stable library item carrying user metadata, selections and relationships to media assets. | W3C provenance model; Media, schema |
| Mermaid | The text-based diagram format used for documentation flowcharts, with top-down flow. | Mermaid flowcharts; Development, releases |
| Metadata observation | A raw or typed source value with its qualified field path, extractor, precision and provenance; conflicting observations remain distinct. | ExifTool tag reference; Ingestion, schema |
| Migration | A versioned transformation or transfer of schemas and records, validated before changing the active authority. | PostgreSQL concurrency; Storage, contracts |
| Model card | A model's published description of capabilities, limitations, training context and acquisition/license information. | Hugging Face model cards; Technology, voice models |
| Model worker | A supervised process or service performing model operations behind a versioned adapter contract. | faster-whisper; Architecture, pipelines |
| Multipart upload | Object publication in separately uploaded parts, whose completion and verification precede catalog admission. | S3 multipart uploads; Storage, architecture |
| Named pipe | A Windows IPC endpoint with access controls, used for local client/runtime communication. | Windows named pipes; Architecture, security |
| Next.js static export | A build producing static website assets from the documentation, requiring no production application server for readers. | Next.js static export; Releases, development |
| Object locator | A backend-neutral reference to a storage profile, key and optional object version, resolved through the artifact adapter. | S3 object storage; Storage, schema |
| openCypher | A property-graph query language family; LadybugDB and ArcadeDB expose qualified dialects rather than assumed interchangeable text. | openCypher project; Graph, technology |
| Outbox | An ordered durable publication queue committed with catalog changes so separate graph writes can recover. | Transactional outbox pattern; Architecture, graph |
| PCM (pulse-code modulation) | A sampled audio representation used for model input, with explicit sample rate, channel layout and sample format. | FFmpeg resampling; Media, technology |
| Pipeline | An immutable revision of stages, adapters, models, provider routes and effective options selected for jobs. | OpenLineage model; Pipelines, schema |
| PostgreSQL | The supported server relational catalog alternative, providing the same operational authority as SQLite. | PostgreSQL concurrency; Storage, technology |
| Precision | The resolution actually supported by a date/time or source observation; date-only values and milliseconds must not imply finer knowledge. | Internet date/time standard; Ingestion, media |
| Presentation timestamp | The media stream time at which content is presented, used instead of assuming a constant video frame rate. | ffprobe documentation; Media, subtitles |
| Provenance | The recorded source, activity, options and relationships that explain how an observation or artifact was obtained. | W3C provenance model; Media, schema |
| Provider | The local service or remote host selected for an adapter, with explicit endpoint and credential references. | OpenAPI concepts; Pipelines, security |
| PyTorch | A tensor and model-training framework used by suitable optional Python workers rather than the shared Go application core. | PyTorch model tutorial; technology, voice models |
| Quantization / int8 | Representing model values with lower precision, including eight-bit integers, to change memory and inference costs under a qualified model profile. | CTranslate2 quantization; technology |
| QuerySpec | insonic's versioned normalized query record, compiled by each graph adapter into backend-specific operations. | OpenAPI concepts; Graph, contracts |
| React | The component-based user-interface library for the desktop UI. | React learning guide; technology, desktop |
| Read-only operation | A query mode constrained to observation through adapter parsing, privileges and engine-specific enforcement. | SQLite SQL reference; Graph, security |
| Recording-time observation | A date/time candidate carrying its literal value, precision, zone, source and assumptions; a separate selection drives calendar placement. | Internet date/time standard; Ingestion, graph |
| Revision | An immutable version of evidence or a recorded change to a selection/association; identifiers retain what a job or result used. | W3C provenance model; Schema, speakers |
| S3-compatible storage | An object-store API role qualified for required operations; provider compatibility does not imply every AWS-specific feature. | S3 object storage; Storage, technology |
| Secret store | Encrypted credential storage through an OS service or an explicit encrypted/session-only fallback, addressed by nonsecret credential IDs. | Secret Service API; Security, storage |
| SemVer (Semantic Versioning) | A major.minor.patch release version describing compatibility; software, docs and the master schema share the insonic release version. | Semantic Versioning; Releases, contracts |
| Snapshot (backup) | A consistent catalog revision paired with an artifact manifest and retained bytes or explicitly dependent external references. | SQLite backup API; Storage, architecture |
| Speaker | A stable catalog identity distinct from a file-local acoustic voice, with names, aliases and revisioned attribution. | pyannote audio; Speakers, schema |
| Speaker audio segment | A source-timed audio interval with stream/channel and acoustic evidence, linked separately to revisioned speaker attribution. | pyannote audio; Speakers, voice models |
| Speaker model | A trained, versioned artifact retaining its originating speaker, immutable dataset snapshot and processing receipt. | PyTorch model persistence; Voice models, schema |
| Spec Kit / work slice | The specification workflow for a bounded observable outcome, with implementation tasks connected to GitHub tracking. | Spec Kit; Development |
| SQL (Structured Query Language) | A relational query language implemented through qualified backend dialects; adapters keep dialect details out of processing logic. | SQLite SQL reference; Storage, graph |
| SQLite | The default local operational catalog, opened with configured transactional and foreign-key behavior. | SQLite transactions; Storage, technology |
| Stream / channel | A selected media stream and its channel layout, retained in source and derivative relationships so useful separation is preserved. | FFmpeg resampling; Media, ingestion |
| Subtitle formats (SRT, WebVTT, ASS, SSA) | Native timed-text formats admitted/rendered by Cueson as subrip, webvtt, ass and ssa, with format-specific fields and limitations. | Cueson schema; Subtitles |
| Term | A user-maintained spelling or context hint supplied with speaker names to capable transcription/reasoning adapters. | faster-whisper; Speakers, pipelines |
| Time map | A recorded mapping from derivative/chunk time to original media time; cuts or concatenation need a piecewise map. | W3C provenance model; Media, subtitles |
| Time zone (IANA) | A named set of historical/local clock rules, resolved with a recorded database version when interpreting origination times. | IANA time-zone database; Ingestion, desktop |
| TLS (Transport Layer Security) | Encrypted transport with peer verification for configured remote providers and database/object-store services. | TLS 1.3 standard; Security, storage |
| Transaction / ACID | An atomic, consistent, isolated and durable database operation; independent artifact, catalog and graph systems do not share one transaction. | SQLite transactions; Architecture, storage |
| Transcript revision | An immutable normalized subtitle document and its provenance, with an explicit current selection. | Cueson schema; Subtitles, schema |
| TypeScript | JavaScript with static type checking, used for desktop UI contracts; runtime inputs still require validation. | TypeScript documentation; technology |
| UTC (Coordinated Universal Time) | The reference time scale for normalized instants and audit times, distinct from media-relative timing and local calendar display. | Internet date/time standard; Ingestion, schema |
| UUID (universally unique identifier) | A portable identifier form used for stable workspace/entity IDs rather than backend-native row or graph addresses. | UUID standard; Schema, contracts |
| Voice | A run-local acoustic cluster that may remain unnamed and searchable independently of a speaker identity. | pyannote audio; Speakers, graph |
| Wails | The Go-based desktop framework joining a native webview UI to the shared application/runtime. | Wails introduction; Technology, desktop |
| WAL (write-ahead log) | A database recovery/concurrency mechanism recording changes before their final placement; SQLite WAL files are not a complete media backup. | SQLite WAL; Storage, architecture |
| whisper.cpp | A native Whisper implementation offering CPU and platform-specific acceleration behind the transcription adapter. | whisper.cpp project; technology |
| Workspace | A logical library and local runtime/cache root connected to independently selected storage, catalog and graph profiles. | S3 object storage; Storage, architecture |