You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: CHANGELOG.md
+19Lines changed: 19 additions & 0 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -4,6 +4,25 @@ All notable changes to this project will be documented in this file.
4
4
5
5
The format is based on [Keep a Changelog](https://keepachangelog.com/), and this project adheres to [Semantic Versioning](https://semver.org/).
6
6
7
+
## [0.10.0] - 2026-04-26
8
+
9
+
### Added
10
+
-**StepAudio 2.5 ASR** — added StepFun batch transcription via HTTP + SSE, including API key/model settings, provider selection, history metadata, and re-transcription support.
11
+
-**Per-app text injection overrides** — Input Settings can now remember recent target apps and choose pasteboard or typing mode per application, so apps with special editor behavior can use their own injection strategy.
12
+
-**LLM connectivity test** — Settings → LLM can send a real correction probe with the active provider and model, then show status, response time, test input, and model output.
13
+
-**History replay injection test** — saved recordings can now be re-transcribed, post-processed, and injected into the current foreground app for end-to-end provider and injection checks.
14
+
15
+
### Changed
16
+
- Expanded VoiceX from eleven to twelve ASR backends.
17
+
- Re-transcription now returns the final post-processed text in addition to ASR and LLM intermediate results, and it respects the original history mode when choosing assistant vs. translation prompts.
18
+
- Clipboard injection and macOS paste shortcuts now use steadier timing to improve reliability in editors that process paste events slowly.
19
+
- The compact HUD now keeps the processing intent chip visible during batch and compact states.
20
+
21
+
### Fixed
22
+
- Prevented the HUD from stealing input focus and preserved the original foreground app context when recording starts.
23
+
- Hardened hotkey session races around rapid start/stop/cancel flows.
24
+
- Empty Qwen sessions that exit because of silence are now handled without leaving stale session state.
Copy file name to clipboardExpand all lines: README.en.md
+10-3Lines changed: 10 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -19,8 +19,8 @@ VoiceX is a cross-platform desktop voice input tool. Its overall pipeline is: re
19
19
-**One hotkey, multiple gestures** — a single global hotkey drives three interaction modes: tap for hands-free dictation, hold for push-to-talk, double-tap to translate.
20
20
-**Real-time HUD overlay** — a lightweight always-on-top display shows live transcription, recording mode, countdown timer, and processing status, and on macOS it follows the active Space when triggered from another desktop.
21
21
-**LLM-powered post-processing** — optionally send ASR output through an LLM for correction, translation, or refinement, with customizable prompt templates and dictionary-aware context.
22
-
-**Smart text injection** — recognized text is pasted into the active app via clipboard (with automatic backup/restore) or simulated typing, seamlessly.
23
-
-**History & statistics** — every dictation is logged with full metadata (duration, device, ASR/LLM model, original vs. corrected text), browsable by date with audio playback and re-transcription.
22
+
-**Smart text injection** — recognized text is pasted into the active app via clipboard (with automatic backup/restore) or simulated typing, with per-app overrides for apps that need a different injection strategy.
23
+
-**History & statistics** — every dictation is logged with full metadata (duration, device, ASR/LLM model, original vs. corrected text), browsable by date with audio playback, re-transcription, and replay injection testing.
24
24
-**Cross-device sync** — a self-hosted sync server keeps history in sync across your machines.
25
25
26
26
## Interaction Modes
@@ -73,6 +73,7 @@ Features:
73
73
-**ASR correction** — fix recognition errors using dictionary context and customizable prompts.
74
74
-**Translation** — translate dictation to English, triggered by double-tap gesture.
75
75
-**Prompt templates** — full control over correction and translation prompts, with `{{DICTIONARY}}` placeholder for hot-word injection.
76
+
-**Connectivity test** — send one real correction request with the active provider and model to inspect latency and output quality.
76
77
77
78
## Dictionary & Hot-Words
78
79
@@ -85,11 +86,17 @@ Features:
85
86
-**Smart punctuation cleanup** — auto-remove trailing punctuation from short sentences (configurable threshold).
86
87
-**Keyword substitution** — regex/exact/contains replacement rules applied before text injection.
87
88
89
+
## Text Injection
90
+
91
+
- Global support for clipboard paste and simulated typing modes.
92
+
- Per-app overrides can be created from recent target apps, so standard text fields can keep clipboard paste while editors that are sensitive to paste events can switch to typing.
93
+
- On macOS, VoiceX records the target app when the hotkey starts, reducing the chance that the HUD or main window changes the injection target.
94
+
88
95
## History & Statistics
89
96
90
97
- Full history grouped by date, with per-record audio playback, copy, and detail view.
91
98
- Side-by-side comparison of original ASR output vs. LLM-corrected text.
92
-
- Re-transcribe any saved recording with a different ASR backend and optional LLM correction to compare providers on the same audio.
99
+
- Re-transcribe any saved recording with a different ASR backend and optional LLM correction to compare providers on the same audio; replay the final text into the current app to test the full end-to-end flow.
93
100
- Failed batch transcriptions are preserved locally with the original audio so you can retry later instead of repeating the whole dictation immediately.
94
101
- Configurable retention policies for text and audio (7 / 30 / 180 / 365 days, or forever).
95
102
- Overview dashboard: total duration, character count, AI correction calls, average dictation speed — aggregated per device.
0 commit comments