One recording. Everything attached.
The original audio remains the permanent source. Transcript, speakers, notes, NovaAI insights, melody analysis, arrangements and exports belong to that same record.
Foundation at a glance
No recordings yet
What are you recording?
Choose the recording purpose and capture source independently. NovaEar can record audio, webcam video, or screen video. The original media is saved first; Speech and Music only run when you request them.
New personal note
Drag your webcam bubble anywhere in the preview.
Upload audio or music
Bring existing recordings, songs, instrumentals, lectures, meetings or interviews into NovaEar. The original file is preserved first; Speech and Music remain independent.
Recording
Transcript
Speakers
Insights attached to this recording
Analysis & arrangements
Your notes for this recording
Export without lock-in
The JSON export contains record metadata and attached artifacts. Original audio/video remains separately downloadable in its captured format.
Transcript
Build in layers. Keep the voice central.
Voice → pitch contour → notes → rhythm → key → arrangement. v34 adds NovaReference style learning and NovaStem preparation while preserving the three-engine architecture, performance instruments, accent-aware Speech and NovaAI context correction.
My Music Sources
Learn from an uploaded instrumental or song
Choose a saved music source and let NovaEar measure its musical character. The resulting profile guides a new arrangement without copying the reference melody or audio.
Vocals ↔ instrumental pipeline
NovaEar now reserves stem artifacts on each music record so a local separation model can attach Vocals and Instrumental stems without creating a fourth core engine.
Create and perform your own instrument
Build a reusable local synth patch, then play it from your computer keyboard or a compatible MIDI controller. Nothing is uploaded.
Inspect before expanding.
Diagnostics, architecture and sound tools are separated from the everyday recording flow.
Diagnostics
Run browser checks plus independent Audio, Speech and Music engine smoke-tests.
Architecture
Review engine ownership, boundaries and the permanent record contract.
Instrument Vault
Audition generated instruments without coupling synthesis code to the UI.
Playable code-generated sounds
Each card directly calls the Music Engine.
Preserved research catalogue
Three engines. Hard boundaries.
The orchestrator may ask an engine to work. No engine is allowed to own another engine’s responsibility.
1 · Audio Engine
Capture, pause/resume/stop, imports, decode/resample, IndexedDB and canonical record schema.
2 · Speech Engine
Worker lifecycle, audio conditioning, adaptive VAD, quality-tier model routing, uncertain-section retry, repetition rejection and quality scoring.
3 · Music Engine
Smoothed pitch contour, note segmentation, key confidence, chord-candidate scoring, multiple instrumental palettes, custom synth patches and Web Audio rendering.
record ├── original audio blob + metadata ├── transcript + timestamped segments ├── speakers[] ├── notes[] ├── novaInsights[] ├── musicAnalysis └── arrangements[]
Existing v25 records are normalised into this schema when read; the database name is deliberately retained so upgrading does not discard the user’s local library.
Engine isolation diagnostics
These tests do more than detect APIs: Audio performs an IndexedDB write/read/delete round-trip, Speech boots and responds through its worker, and Music verifies its analysis/arrangement contract. The ASR model itself still loads only when transcription is requested. v31 keeps Quick, Balanced and Best tiers, adds accent-aware routing plus NovaLexicon polishing, first-class uploaded music, computer/MIDI instrument performance, and safer voice/backing preview mixing while preserving hard engine boundaries.