FluidVoice

FluidVoice 1.6.11

摘要（非完整翻译）

新增本地会议录制 FluidMeet、全局搜索、个性化仪表盘、语音发送和可定制录音浮层。扩展发音训练、英语 Mini/Pico 模型、模型更新及会议转录 API；改进听写、文本交付、麦克风和会议恢复。会议摘要暂时禁用。

重要说明：会议摘要暂时禁用。Mini 和 Pico 为英语专用模型；发音训练仍属实验功能。现有用户保留 Whisper 的 Automatic 语言检测。听写声音的 Independent Volume 已移除，提示音不再改变 Mac 系统音量。

官方来源: https://github.com/altic-dev/FluidVoice/releases/tag/v1.6.11

官方原文（完整） / 完整英文版本

FluidVoice v1.6.11

Changes

• Introducing FluidMeet: record meetings locally with live captions, speaker labels, and transcripts.
• Introducing Global Search: find history, transcripts, chats, dictionary entries, cleanup styles, and settings across the app.
• Redesigned the app with clearer navigation, improved onboarding, better layouts for smaller windows, and a What’s New popup to discover each update’s highlights.
• Made Custom Dictionary more intuitive, with experimental pronunciation training that learns how you say words—not just text replacements.
• Added Spoken Send: use a configurable spoken phrase to send your completed dictation, with safeguards for terminal apps.
• Added a personalized Dashboard with recent dictations, quick actions, shareable stats, and Fluid Intelligence highlights.
• Customize the recording overlay with new finishes, colors, transparency, edge lighting, and a voice-responsive waveform.
• Refreshed Fluid Intelligence model browsing and provider setup with model discovery, clearer status, and per-model verification.
• Added configurable idle unloading to free memory when Fluid Intelligence is not in use.
• Added optional Hold Space to Mute during dictation, with muted indicators in the bottom and notch overlays.
• Added Fluid Speech Mini and Pico: smaller English-only models with custom words and pronunciation training.
• Update compatible speech-model weights without updating the app, with resumable verified downloads.
• Pin transcripts in History, play individual meeting transcript segments, and double-click meeting text to correct it.
• Added a local API to list meetings and retrieve transcripts.
• Added Whisper language selection while preserving Automatic detection for existing users.

Fixes

• Fixed AI enhancement with newer Claude models, including Haiku 5.5.

• Improved button click areas, removed duplicate logging, and made cancelled queued meeting-model loads return promptly.

• Improved dictation responsiveness and long Parakeet recordings by reducing repeated processing and unnecessary waits.

• Expanded local transcription API uploads to 500 MB and removed the previous fixed transcription-duration limit.

• Added support for WhatsApp .opus/.oga voice notes and additional audio formats in File Transcription.

• Improved feedback validation and delivery confirmation while preserving drafts when sending fails.

• Kept completed dictation tied to the destination and cleanup choices captured when recording stops, even if you switch apps during processing.

• Improved clipboard restoration and text delivery, with optional failure notices and filename preservation when oversized clipboard data is skipped.

• Improved media pause and resume during dictation. Removed Independent Volume for dictation sounds so cues no longer change the Mac's system volume.

• Preserved the complete raw transcript when a long Fluid-1 request exceeds its supported context.

• Honored Send Custom Prompt Only for dictation-shortcut overrides and kept transcript text separate from cleanup instructions in supported requests.

• Prevented Basic Instant Replacements from influencing similar-sounding words through Custom Words Boosting.

• Improved pronunciation matching for longer recordings, preserved possessive endings, and prevented corrected words from being expanded twice.

• Kept ignored microphones excluded after reconnecting and improved microphone startup and device-change handling.

• Prevented shortcuts from triggering while the Mac is locked and improved mouse-shortcut handling.

• Fixed the hidden Dock icon returning when reopening an already-running app.

• Preserved accepted Edit Mode rewrites when closing the panel and restored the original field before replacing its text.

• Kept search results current after startup and backup restoration, and hid history audio actions when recordings are unavailable.

• Cleared stale fetched-model selections when provider connection details change.

• Improved meeting audio capture, speaker-name consistency, interruption recovery, and coordination between meeting tasks.

• Improved recording-overlay sizing and label fit.

Notes

• Meeting summaries are temporarily disabled while accuracy is improved.

Contributors

Thanks to @wallneradam, @grohith327, @juanmartitegui, @HiroProt, @so1omon563, @Ghatage, @postoso, @shreeraman96, @argus-vault, @Lstarsky0, @nishantkumar1292, @aniansh-redacto, @bunnysayzz, @chntllh, @kirolos-esmat, and @alexmodon for fixes and improvements.