Testing 8 AI voice input methods: Qwen input wins on accuracy, WeChat on speed
卡尔的AI沃茨 · wechat · 2026-10-10
The author rounded up the most-requested AI voice input methods and, combined with a previous round (Typeless, Shandianshuo, NetEase Bagshuo, Doubao), tested eight in total using the same mixed Chinese-English script with proper nouns, all on free tiers.
This round: WeChat Input excels at cross-device copy-paste but feels barely AI-powered, transcribing verbatim; Qwen Input has the deepest AI rewriting and the highest accuracy of all eight, with two-way PC-phone sync at a small speed cost; open-source OpenLess (MIT, local-first) lets you configure any ASR+LLM combo, supports custom dictionaries and per-scenario style presets — about 30 minutes of setup gets you near paid-product quality; MetaSota's input never launched successfully on Mac.
The author's final setup: Qwen for mobile and cross-device sync, Bagshuo (50k chars/week) for everyday use, and Typeless (shrinking quota) for speed-critical long-form writing. Voice input has split into "faithful transcription" vs. "understanding rewrite" — the depth of AI intervention spans the gap between recording what you said and understanding what you meant.
More from Apps
- OmarchyTV beta brings DHH's Omarchy-style launcher to Android TV — alexcovo_eth · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- Capy 0.5.0 lets Claude Pro/Max subscribers reuse their plan quota, adds OIDC cloud access — garrytan · 2026-10-23
- ChatGPT + HeyGen MCP turn job listings into video cover letters in your face and voice — HeyGen · 2026-10-23
- User recovers $3,000 in unrefunded returns using ChatGPT's agent mode — dinasaur_404 · 2026-10-23
- Amazon blocks AI shopping agents while Walmart opens its site to them with spend controls — venturetwins · 2026-10-23