Crate phostt 

phostt

Local speech-to-text powered by Zipformer-vi RNN-T — on-device Vietnamese speech recognition via ONNX Runtime. No cloud APIs, no API keys, full privacy.

Quick start

use phostt::inference::Engine;

let engine = Engine::load("~/.phostt/models")?;

// File transcription
let text = engine.transcribe_file("audio.wav")?;

// Streaming recognition
let mut state = engine.create_state(/* diarization_enabled: */ false);
let segments = engine.process_chunk(&audio_16khz, &mut state)?;

Modules

• inference — ONNX inference engine, streaming state, audio utilities
• error — Typed error types (PhosttError)
• protocol — WebSocket JSON message types
• server — WebSocket server entry point
• model — Model download and management

Modules

error

Error types for the phostt public API.

inference

ONNX Runtime inference engine for Zipformer-vi RNN-T.

inspect

ONNX session introspection for the phostt inspect debug subcommand.

model

Model bundle download and management.

protocol

WebSocket protocol messages for phostt.

server

HTTP + WebSocket server that accepts audio and streams transcripts.