Dev#4
Open
Uzaaft wants to merge 9 commits into
Open
Conversation
Configure local Xcode builds to use ad-hoc signing when no developer team is set. Previous local app rebuilds could get a changing or missing signature, which made macOS privacy permissions less stable during development. The build helper now passes explicit manual signing settings for the regular macOS app build path. Archive builds keep their existing behavior.
Add registry entries for the English-only tiny, base, small, and medium Whisper models. The model list previously exposed only multilingual variants, even though the download source ships smaller language-specific files. The descriptors use the same runtime and capabilities as the existing Whisper entries while pointing at the .en filenames and downloads. Larger models remain multilingual-only because no matching English-only files are available.
Let the runtime adapter update decoder prompts and VAD parameters after model load. Previously those values were fixed at initialization, so app-level dictionary prompts and input-device-specific VAD buckets required a reload. The adapter now owns an optional initial prompt and routes prompt and VAD updates to the whisper.cpp backend. Transcribe calls read the current values when building the request.
Add CLI commands for collecting labeled audio snippets and replaying them through tuning presets. Previously tuning VAD and gain behavior required manual recordings and ad hoc transcription runs. The snippet command records WAV plus JSON metadata, while the tune command loads a snippet corpus, applies presets, and reports WER summaries. The debug recording path reuses the same terminal recording loop.
Track capture-time signals needed for better transcription diagnostics and preprocessing. The audio buffer previously exposed only immutable samples and an RMS level, which was not enough to detect dead routes or normalize quiet speech. Audio capture now tracks trailing exact-zero samples, exposes a stopped-recording mutable sample view, estimates noise floor, and peak-normalizes quiet input. A small CoreAudio helper classifies the default input device for Bluetooth-specific warnings and tuning.
Add first-run permission onboarding and a separate non-fatal warning surface in the macOS app. Previously permission recovery was scattered across startup behavior, and advisory conditions reused error presentation. The app now coordinates microphone and accessibility state, opens the dashboard when setup is needed, and lets users re-run onboarding from settings or the menu. Warning text is presented separately from fatal errors so recording status is not overwritten.
Add local helpers for decoder vocabulary biasing and nonsense transcript filtering. The app previously had no typed owner for a user dictionary and no reusable heuristic for dropping obvious silence hallucinations. The dictionary store loads a bounded plain-text phrase list and builds a prompt suitable for the decoder. The sensibility guard rejects known bad outputs and repeated-token loops without invoking another model.
Add the runtime API fields and app-state plumbing for warning callbacks, whisper-mode VAD tuning, dictionary prompts, adaptive trimming, peak normalization, and local nonsense filtering. The previous app path used fixed VAD and trim behavior with only fatal errors; the new path carries non-fatal warnings separately and prepares captured audio before transcription.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
No description provided.