Push-to-talk dictation for Android apps and terminal workflows

Name: Push-to-talk dictation for Android apps and terminal workflows
Availability: InStock
Author: pol_avec

by pol_avec·Mar 12, 2026·2 points·0 comments

Visit Project View on HN

AI Analysis

●●SolidSolve My ProblemCozy

Floating overlay dictation that keeps your keyboard - Termux command mode converts speech to CLI.

Strengths

•Accessibility Service insertion works across apps without replacing your existing keyboard.
•Local sherpa-onnx transcription keeps audio on-device with optional OpenAI cleanup.
•No backend server - cloud requests go directly from phone to OpenAI with user's API key.

Weaknesses

•Voice dictation category is crowded with Google's built-in input and dedicated apps.
•Zero stars on GitHub suggests early launch with limited real-world validation.

Post Description

I built this because MacWhisper is not available on Android and voice typing on Android is pretty bad. Moreover Gemini does not allow you to edit transcripts before they are auto-sent.

I like my SwiftKey keyboard though, so I did not want to replace that. So the only way was to make a floating push-to-talk button on top of any app.

You tap the overlay, speak, tap again, transcribe, and insert text into the currently focused field.

It supports local on-device transcription, cloud transcription with your own OpenAI key, and optional post-processing/cleanup for punctuation, formatting, prompts, commands, etc.

A nice use case for me has been Termux / terminal workflows on Android. You have a "dev mode" where you can just say "command mode" and anything after it will be converted into a proper CLI command.

The app is open source. No backend — in cloud mode requests go directly from the phone to OpenAI using the user's own API key.

Repo: https://github.com/kafkasl/phone-whisper APK: https://github.com/kafkasl/phone-whisper/releases