Shall We Talk

Product Introduction

1. Definition: Shall We Talk is an open-source (GPL-3.0) native iOS and macOS application suite for advanced voice dictation and meeting transcription. It functions as a custom keyboard, Shortcuts action, and system-wide hotkey tool.
2. Core Value Proposition: It exists to provide privacy-focused, high-accuracy speech-to-text that preserves the user's original wording and tone, including complex linguistic patterns like Chinese-English code-switching, unlike AI rewriters. It turns natural speech into clean, usable text directly at the cursor or into structured meeting notes.

Main Features

1. Multi-Platform Voice Keyboard: A custom iOS keyboard extension that delivers recognized text directly to the active text field's cursor position. It includes advanced Chinese input methods like full-pinyin, T9, and initials lookup for seamless multilingual dictation.
2. System Automation Integration: Offers deep integration with iOS Shortcuts for background dictation (returning text for clipboard actions) and macOS global hotkeys for instant dictation with focused-field insertion or clipboard fallback, enabling complex voice-driven workflows.
3. Intelligent Meeting Recorder: Captures multi-speaker audio with automatic voice activity detection (using local Silero VAD), handles interruptions, and produces speaker-labeled transcripts alongside AI-generated structured summaries, with audio persisted for review.
4. Context-Aware Cleanup Engine: A proofreading layer that removes filler words ("um," "ah") and misheard terms from speech-to-text output without altering the user's original sentence structure, tone, or mixed-language phrasing, ensuring clean text that still sounds like the user.
5. Bring-Your-Own-Keys (BYOK) Architecture: The open-source development repository requires users to supply their own speech recognition (e.g., OpenAI Whisper, Google Speech-to-Text) and Large Language Model (e.g., GPT-4, Claude) API keys, ensuring data control and eliminating vendor lock-in for the backend services.

Problems Solved

1. Pain Point: Inaccurate and tone-deaf dictation. Standard speech-to-text often mishears words and outputs robotic, rewritten text that loses the speaker's unique voice and fails with code-switching, common among bilingual professionals.
2. Pain Point: Fragmented meeting documentation. Manually transcribing meetings is time-consuming, and most automated tools lack accurate speaker diarization and produce unstructured, hard-to-scan text blocks.
3. Target Audience: Bilingual professionals (e.g., analysts, academics, managers in global firms), journalists, content creators, developers with RSI, and anyone needing hands-free, accurate text input across iPhone and Mac.
4. Use Cases: Dictating emails and reports in mixed Chinese-English, recording and summarizing client discovery calls, live-captioning brainstorming sessions, and providing accessible input for users with mobility impairments.

Unique Advantages

1. Differentiation: Unlike cloud-only dictation apps or AI writing assistants that paraphrase, Shall We Talk acts as a "cleanup" layer, prioritizing fidelity to the user's original speech. Its open-source, BYOK model contrasts with closed-source SaaS alternatives, offering full transparency and data control.
2. Key Innovation: The combination of a system-level input method (keyboard/hotkey) with a sophisticated, user-tone-preserving cleanup engine specifically optimized for Chinese linguistic contexts. Its shared Swift core for iOS and macOS ensures feature parity and a unified codebase.

Frequently Asked Questions (FAQ)

1. Is Shall We Talk a free dictation app? Shall We Talk is open-source software (GPL-3.0) that is free to use and modify, but it requires you to provide and pay for your own API keys for speech-to-text and LLM services (like OpenAI or Google Cloud), which incurs separate costs based on usage.
2. How does Shall We Talk handle privacy compared to Siri or Google Assistant? Since it uses a bring-your-own-keys model, your audio and text data are sent directly to the API providers you configure (e.g., your own OpenAI account). The app itself does not have a centralized server, giving you control over your data pipeline and compliance.
3. Can I use Shall We Talk to dictate into any application on my iPhone? Yes, the custom keyboard works in any app that accepts text input. However, due to iOS sandboxing, the keyboard cannot directly insert text into other apps' fields programmatically; for full automation, you use the Shortcuts action which returns text to the clipboard for pasting.
4. What is needed to build and install Shall We Talk from source? You need a Mac with Xcode 26+, XcodeGen, and a configured Apple Developer account to sign the apps. You must clone the repository, configure your own API keys and bundle identifiers, and build the iOS and macOS targets yourself, as it is not available on the App Store.
5. Does the meeting transcription feature work offline? The voice activity detection (VAD) runs locally on-device using Silero VAD to segment speech. However, the actual speech-to-text conversion and summary generation require an internet connection to your configured cloud APIs (e.g., OpenAI Whisper, GPT-4).