@sinhala-typer/voice-typing

A generic Web Speech API + Web Audio voice-typing wrapper. Not Sinhala-specific — despite the package scope, lang is fully caller-configurable and this has zero dependency on @sinhala-typer/typing-engine or any other package in this monorepo. It happens to ship alongside a Sinhala typing tool; it isn't about Sinhala.

What it adds on top of the raw SpeechRecognition API:

• Diffed edits, not wholesale replacement. The Web Speech API repeatedly re-reports a growing/refining transcript for the same utterance ("hell" → "hello" → "hello world"). This computes the minimal { deleteCount, insertText } delta between what you last showed and the new transcript, so you never have to blindly delete-and-retype the whole phrase on every update.
• A volume meter (AnalyserNode-based, not the deprecated ScriptProcessorNode) for "is it hearing me" visual feedback — a second, independent microphone stream, since SpeechRecognition exposes no way to observe its own internal audio capture.
• A start-timeout safeguard. In some environments recognition.start() returns normally but the browser never fires any event — not onstart, not onerror. Left alone this hangs a UI in "listening" forever; this package gives up and reports a clean error after 8 seconds if the browser never confirms it actually started.
• Structured errors, not raw browser error strings.

Install

npm install @sinhala-typer/voice-typing

Usage

import { createVoiceTyping } from "@sinhala-typer/voice-typing";

const voice = createVoiceTyping({
  lang: "en-US", // defaults to "si-LK" if omitted
  onEdit: (edit) => {
    // apply edit.deleteCount / edit.insertText to your own text buffer,
    // e.g. via @sinhala-typer/typing-engine-dom's TypingController.applyExternalEdit()
  },
  onStateChange: (state) => console.log(state), // "idle" | "listening"
  onVolumeChange: (level) => console.log(level), // 0-1
  onError: (error) => console.error(error.code, error.message),
});

if (voice.isSupported) {
  voice.start();
  // later:
  voice.stop();
}

Browser support

Requires SpeechRecognition/webkitSpeechRecognition. Chrome, Edge, and Safari support it; Firefox has never shipped it — this is a real, permanent platform gap, not something this package works around. Check isSupported and hide/disable your UI accordingly rather than assuming it's always available.

API

• createVoiceTyping(options) → VoiceTypingController
  • isSupported: boolean
  • start() / stop()
• VoiceTypingOptions:
  • lang?: string — BCP-47 locale, defaults to "si-LK".
  • onEdit(edit: Edit) — required. Edit = { deleteCount: number; insertText: string }.
  • onStateChange?(state: "idle" | "listening")
  • onVolumeChange?(level: number) — 0-1, best-effort (silently omitted if the browser denies the metering microphone stream).
  • onError?(error: VoiceTypingError) — VoiceTypingError.code is one of "not-allowed" | "no-speech" | "audio-capture" | "unsupported" | "unknown".

License

MIT