Repository navigation
Add isolated Sopro V2 Turbo CPU speech candidate - #44
Conversation
|
Warning Review limit reachedNext included review available in 30 minutes. View limit detailsLimit details: You’ve used the included review currently available. You've used all free OSS reviews for now. Wait for the free limit to reset to keep reviewing this public repository. Review configuration: ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (12)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
Final candidate code passed the opt-in packaged Mac hardware test after the self-contained WAV writer change: first audio 1.11 s, total model load/inference 1.72 s, 3.02 s 24 kHz output. The standalone CLI smoke also passed. These are short-fixture, post-install measurements; quality and other platforms remain unqualified. |
|
All five platform bundle jobs and the Go/docs validation jobs are green on the final commit. The opt-in Apple Silicon speech smoke also passed locally. Requesting code review for merging the isolated candidate; the public picker remains unchanged and issue #36 stays open for cross-platform/quality qualification. |
Summary
Adds an isolated, experimental Sopro V2 Turbo CPU speech adapter and exact six-file candidate manifest. It is intentionally not activated in the public catalog. Apple Silicon macOS is the only runtime-enabled platform until Windows/Linux and voice-quality qualification. A local 5–20-second voice reference plus explicit
--voice-consentare required.Research and test date: 2026-09-18.
Validation
f747f9edfb7b0233a3b7105af3a75603a7213d26, Apache-2.0 access/license, sizes and full SHA-256 for all six files (~605 MiB total).sopro==2.2.0CPU runtime works offline with a locally synthesized, 11.99-second test voice; valid 24 kHz mono WAV.tapioca pull sopro-v2-turboreverified all cached files andtapioca ttsgenerated a valid 3.10-second WAV. With the environment installed, first audio was 1.13 seconds and total model load/inference 1.78 seconds for this single short English fixture.go test ./...and Python validation tests passed.Remaining before public support / issue closure
Perceptual intelligibility and voice similarity in all four languages, malformed/alternate reference handling, long text, cancellation cleanup, peak memory, and Windows/Linux CPU paths. The first-run Python install is separate from the above generation timing. Please keep #36 open; this is a candidate, not a release claim.
Sources
Related: #36.