Sayboard は 3 つの認識エンジン + GPT-4o で、あなたの声をきれいで構造化されたテキストに変換します。Fn を長押しして話し、離すだけ。
Apple Speech + ローカル Whisper large-v3-turbo + クラウド Whisper が並列動作。クロス検証で単一エンジンでは達成できない精度を実現。
すべての音声認識結果を GPT-4o が校正——同音異義語の修正、句読点の追加、段落の構造化、箇条書きの自動フォーマット。
Fn を長押しして話し、離すとペースト。2 秒以内にカーソル位置にテキストが表示。アプリの切り替え不要。
中国語、英語、日本語、韓国語、フランス語、スペイン語など 90 以上の言語に対応。言語の混在も自由自在。
Whisper large-v3-turbo はあなたの Mac 上で完全に動作。音声はローカルに保持。基本認識にクラウドは不要。
Fn キーによるシステムレベルの音声入力。Slack、VS Code、LINE、Chrome、メモ、ターミナルなど、テキスト入力可能なすべてのアプリで使用可能。
Mac のどこでも Fn キーを長押しして録音開始。アプリを開く必要はありません。
普段通りのペースで話してください。3 つのエンジンが同時に音声を文字起こしし、リアルタイムプレビューが表示されます。
Fn を離すと、GPT-4o があなたの言葉をきれいで構造化されたテキストに変換し、カーソル位置に即座にペーストします。
| Sayboard | Typeless | MacWhisper | |
|---|---|---|---|
| Primary use | Real-time voice input | Real-time voice input | File transcription |
| Recognition engines | 3 (cross-validated) | 1 | 1 |
| AI post-processing | GPT-4o (auto formatting, correction) | AI auto-editing (filler removal) | GPT/Claude (BYO API key, Pro) |
| Auto structuring | Bullet points, numbered lists, paragraphs | Basic auto-formatting | No |
| Self-correction | Yes ("delete that" removes prior text) | Yes (mid-sentence correction) | No |
| On-device model | Whisper large-v3-turbo | On-device (model unspecified) | Whisper (Tiny to Large-V3, Parakeet v2) |
| Price | $4.2/mo (billed yearly $49.9) or $9.9/mo · Free tier available |
$12/mo (billed yearly) or $30/mo · Free tier available |
€64 one-time Free tier with basic models |
無料で始めて、必要に応じてアップグレード。
Most voice typing apps rely on a single recognition engine. Sayboard runs three in parallel — Apple Speech for real-time preview, on-device Whisper large-v3-turbo for high-accuracy local transcription, and cloud Whisper for cross-reference. When engines disagree, Sayboard picks the most accurate result. This cross-validation approach catches homophone errors, brand names, and technical terms that any single engine would miss.
After recognition, every result passes through GPT-4o for intelligent refinement. This isn't simple spell-check — Sayboard's AI understands your context from recent conversation history, corrects same-sound character errors (especially critical for Chinese), adds proper punctuation, and automatically structures long dictation into paragraphs, numbered lists, or bullet points. Say "delete that" or "that's wrong" and Sayboard removes the incorrect part, keeping only your intended text.
Sayboard works as a system-level input method triggered by holding the Fn key. No need to switch apps, open a window, or click a button. Press Fn in any text field — Slack, VS Code, WeChat, Chrome, Notes, Terminal, or any other app — speak naturally, and release. Your polished text appears at the cursor in under 2 seconds. It's the fastest way to turn speech into text on a Mac.
Sayboard excels at mixed-language dictation — a common scenario for bilingual users. Speak Chinese with English brand names, technical terms, or abbreviations mixed in, and Sayboard handles it seamlessly. The AI understands that "赛博的" likely means "Sayboard", "A派" means "API", and preserves correct English capitalization (GitHub, iPhone, WebSocket). No need to switch language modes mid-sentence.
Sayboard bundles the Whisper large-v3-turbo model (1.6GB) directly on your Mac. Basic recognition runs entirely on-device with no internet connection required — your audio never leaves the machine. Cloud recognition is optional, used only to cross-validate for higher accuracy. For users handling sensitive or confidential content, Sayboard's local-first architecture ensures complete privacy.
Sayboard doesn't just transcribe — it formats. When you dictate multiple points ("First... Second... Third..."), the output is automatically organized into a numbered list. Long monologues are broken into readable paragraphs with proper line breaks. Filler words like "um", "uh", "you know" are silently removed. The result reads like carefully typed text, not a raw transcript.
Every recognition is saved to your history with timestamps, so you can review, copy, or reuse past dictations anytime. History syncs across devices via your Sayboard account. Your recent history also feeds into GPT-4o as context, helping the AI understand your current topic and improve accuracy for domain-specific vocabulary over time.
Sayboard provides separate optimized builds for Apple Silicon (M1/M2/M3/M4) and Intel Macs. The website automatically detects your chip and offers the right download. On Apple Silicon, the on-device Whisper model leverages the Neural Engine for faster-than-realtime transcription. Intel Macs are fully supported with the same feature set. Requires macOS 14 (Sonoma) or later.
Yes. Sayboard offers a free tier with 20 recognitions per day, including all 3 recognition engines and AI post-processing. Pro plans start at $4.2/month (billed yearly at $49.9) for unlimited recognitions.
Apple's built-in dictation uses a single engine and outputs raw text. Sayboard uses 3 engines for higher accuracy and passes every result through GPT-4o, which corrects errors, adds punctuation, structures paragraphs, and formats lists — producing polished text that's ready to send.
Partially. The on-device Whisper model and Apple Speech work without internet. However, cloud Whisper cross-validation and GPT-4o post-processing require an internet connection for best results.
Sayboard supports any Mac running macOS 14 (Sonoma) or later, including both Apple Silicon (M1, M2, M3, M4) and Intel processors. Separate optimized builds are available for each architecture.
Absolutely. Sayboard is specifically optimized for mixed Chinese-English dictation. Speak Chinese naturally with English terms mixed in — brand names, technical vocabulary, abbreviations — and Sayboard's AI correctly identifies and preserves each language with proper capitalization.