ENGINEERING INDEX / ENTRY 20
CASE 20 / AUDIO / MODEL ROUTING
The voice choice must reach the right engine.
Reader may expose more than one narration engine. A voice selection needs to map to the matching engine, and cached audio must distinguish those voices.
Source project: Twilight ReaderExploratory router checkReviewed September 25, 2026
How does Reader avoid handing a selected voice to the wrong speech engine or mixing cache entries?
02 / THE SYSTEM
Where the handoffs happen.
A stub-engine test selects the F5-labeled voice, confirms only that engine was called, and verifies its cache key differs from the other voice for the same words.
- 01Select voice
- 02Resolve engine
- 03Synthesize
- 04Key cache by engine identity
What the record earns.
One stub two-engine routing and cache-identity fixture passed; no real audio was compared.
- Observed receipt
- On September 25, 2026, the Reader TTS-router test passed within a seven-test extraction/router batch.
- Source and scope
- Twilight Reader SpeechRouter, AudioCache, and tests/test_reader_tts_router.py; Reader source/tests untracked in dirty checkout with base HEAD efd4149.
04 / THE LIMIT
What this does not prove.
Stub audio verifies routing and identity, not real voice quality, GPU performance, rights, or a full-book render.
Where the work goes next.
Run a short rights-cleared two-voice audio comparison and verify cache reuse and invalidation with real engine versions.