xsoulspace_inference_web_speech_recognition

Chromium-only Web Speech API (SpeechRecognition) implementation of xsoulspace_inference_core for speech-to-text.

Supported task

  • InferenceTask.speechToText

Unsupported tasks return task_unsupported.

Platform matrix (v1)

  • Chromium-family browsers (Chrome, Edge, Opera, Brave): supported
  • Safari / Firefox / non-web runtimes: return task_unsupported

Browser constraints

  • Uses Web Speech API constructor from window.SpeechRecognition or window.webkitSpeechRecognition.
  • v1 returns one final transcript per infer(...) call (no streaming API).
  • URL inputs must be browser-loadable (https:, blob:, data:).
  • File URL and bytes modes use <audio>.captureStream() audio track + recognition.start(audioTrack).
  • If start(audioTrack) is unsupported at runtime, inference fails with task_unsupported and details reason audio_track_start_unsupported.

Install

dependencies:
  xsoulspace_inference_web_speech_recognition:
    path: ../xsoulspace_inference_web_speech_recognition
  xsoulspace_inference_core:
    path: ../xsoulspace_inference_core

Usage

import 'package:xsoulspace_inference_core/xsoulspace_inference_core.dart';
import 'package:xsoulspace_inference_web_speech_recognition/xsoulspace_inference_web_speech_recognition.dart';

final client = WebSpeechRecognitionInferenceClient();

final micResult = await client.infer(
  InferenceRequest.speechToText(
    audioInput: const InferenceAudioInput.microphone(
      mimeType: 'audio/webm',
    ),
    metadata: const <String, dynamic>{'language': 'en-US'},
  ),
);

final urlResult = await client.infer(
  InferenceRequest.speechToText(
    audioInput: const InferenceAudioInput.filePath(
      filePath: 'https://example.com/sample.wav',
      mimeType: 'audio/wav',
    ),
  ),
);

if (!urlResult.success) {
  throw StateError('${urlResult.error?.code}: ${urlResult.error?.message}');
}

final transcript = urlResult.data?.transcript;
final normalized = urlResult.data?.normalizedTranscript;

Error mapping

  • Permission/service blocked (not-allowed, service-not-allowed): task_unsupported
  • Invalid/capture/input issues (audio-capture, invalid URL/bytes/track): audio_input_invalid
  • Runtime engine/network/language failures (network, language-not-supported, internal runtime failures): engine_unavailable

Generation

Raw interop files are generated only from the pinned declaration snapshot:

  • Source snapshot: tool/generated/web_speech_recognition.generated.d.ts
  • Generated raw output: lib/src/raw/web_speech_recognition_raw.g.dart

Commands:

cd pkgs/xsoulspace_inference_web_speech_recognition
just generate
just generate-check

Verification

cd pkgs/xsoulspace_inference_web_speech_recognition
just analyze
just test
just test-web

Libraries

raw
xsoulspace_inference_web_speech_recognition