Yan
Start voice input with a shortcut without leaving the app you are using, so your train of thought stays uninterrupted.
Watch the text take shape as you speak, keeping your words in step with your thoughts.
Use voice input naturally across writing, editing, and communication apps.
Set your shortcut once, then start voice input whenever you need it.
Choose a convenient shortcut that does not conflict with your other apps.
Press the shortcut where you want to write to begin voice input.
Focus on what you want to say. Yan turns your voice into words where you are writing.
VOICE INPUT, RE-ENGINEERED
Next-generation speech models such as Qwen3-ASR understand expression; Yan's own inference engine makes them responsive and reliable on your computer
Yan supports multiple local speech models, so you can choose based on language, quality, and device performance
01 / THE MODEL
For Chinese dialects, mixed Chinese and English, varied accents, and specialized terms, next-generation speech models offer stronger language understanding; Yan lets you choose from multiple local models for different situations
02 / THE ENGINE
Real-time streaming, hardware backend selection, and resource scheduling determine whether a large model can become an everyday input tool; Yan's inference engine turns model capability into a responsive desktop voice-input experience
CURRENT ASR LIMITS
QWEN + YAN ENGINE · MODEL + ON-DEVICE ENGINE
General-purpose ASR often miswrites people, brands, and domain terms
Qwen3-ASR jointly decodes audio and language information; dictionary rules correct specialized terms
Partial results are revised as more speech arrives
An early decode lowers time to first text; the refiner corrects and merges stable text
Language-specific pipelines can switch incorrectly when one recording spans multiple languages or songs
A full-film test continuously recognized English, Spanish, German, and English-language songs
Traditional runtimes prioritize NVIDIA, leaving AMD and Intel GPUs underused
Metal runs 1.7B F16 smoothly on M3, while optimized Vulkan runs 1.7B Q8 smoothly on Intel Xe integrated graphics
VOICE INTELLIGENCE, ON-DEVICE
From models and inference to the desktop input experience, Yan connects every step from voice to text
Transcription quality and speed depend on the selected model, language, accent, recording environment, and device performance
Yan's on-device speech engine handles recognition locally, so your voice and transcripts do not need to pass through the cloud.
Speech recognition runs on your device, so your audio does not need to be uploaded to the cloud.
Yan does not use your voice or transcripts to train models.
Your transcript history stays on your device. Yan does not store a copy on its servers.
From speech models to transcription history, the essentials stay visible and under your control.
Compare language support, speed, and resource use, then switch when your needs change.
Review, search, copy, and export transcriptions from one place on your device.
Compare language support, speed, and resource use, then switch when your needs change.
Review, search, copy, and export transcriptions from one place on your device.
From writing and communication to everyday tasks, Yan turns natural speech into clear text
Draft directly in documents and editors, so your words keep up with your thoughts.
Draft messages and emails quickly without repetitive typing.
Add content to documents, spreadsheets, and collaboration tools more efficiently.
Draft replies, explanations, and internal updates while staying focused on the conversation.
Capture thoughts as they come and keep them as text you can refine later.
The icons show common use cases. Compatibility depends on each app and text field supporting system text input.
ADDITIONAL CAPABILITY
Alongside real-time voice input, Yan can batch-transcribe common audio and video files on your device, ready to review, copy, or export.
Try every feature free for 7 days. When you're ready, buy once and keep using Yan with no subscription.
Perpetual use on one device
14-day refund policy · View policy
No payment required. The trial includes everything in the full license.
