Anti-Vocale

Anti-Vocale transcribes voice messages from WhatsApp, Telegram, and other messaging apps directly on your Android device. All processing happens on-device: your audio never leaves your phone. It works on recorded calls and meetings too: recordings of any length are split and stitched automatically (up to two hours on the streaming path), a failed run keeps the transcript it produced, and long transcripts can be exported as subtitles (SRT, VTT) or timestamped text, with sentence-level timing on streaming models. BUILT-IN MODELS - Parakeet TDT (recommended): fast, 25 European languages. Two variants: high-quality (862MB) or compact (464MB) - Whisper Turbo: 99 languages, balanced speed and quality - Whisper Distil Italian: best Italian quality, tuned for voice messages - Whisper Medium / Small: more accuracy or smaller download - Nemotron 3.5 streaming: text appears in real time while it decodes, 45 languages - Qwen3-ASR: compact, 59 languages - GigaAM v3: Russian, very small and fast - Gemma 4 / 3n (LiteRT): multimodal AI with customizable prompts BRING YOUR OWN MODELS Import custom transcription models from a folder on your phone or from a link: Whisper fine-tunes, CTC models, SenseVoice and other sherpa-onnx architectures. Files are integrity-checked on import, split-file models are supported, and every imported model gets its own share target. Ideal for languages the built-in models don't cover. FEATURES - Share voice messages from any app via the Android share sheet - Smart notifications: one-tap copy, share back, page through long transcripts without opening the app - Auto-copy to clipboard and auto-save transcripts to a folder you pick (Drive, Syncthing, ...) - Video file transcription (audio track extracted) - Real-time progressive text with streaming models - Per-app notification preferences - Transcription log with search and swipe actions - Silence stripping and parallel chunk processing for speed - Neural acceleration (NNAPI) on supported devices, with automatic CPU fallback - Tasker integration for automation - Scheduled folder watch: transcribe new audio files in folders you choose automatically (on Android 11+, even a messaging app's media folder) - Remote offload (experimental, off by default): delegate long transcriptions to your own OpenAI-compatible server PRIVACY No network needed after download, no accounts, no data collection, no ads, no tracking. By default the audio never leaves your phone; the only exception is the optional Remote offload feature, which sends audio solely to the server address you configure and stays off unless you turn it on. OPEN SOURCE Apache 2.0. Source code: https://github.com/RisorseArtificiali/anti-vocale REQUIREMENTS Android 8.0+ | 4GB+ RAM recommended | 326MB-1GB storage per model NOTE: on some phones with aggressive battery savers (Realme, OnePlus, Vivo), open the app once after sharing audio so the transcription can finish.