108 lines
3.3 KiB
Dart
108 lines
3.3 KiB
Dart
import 'dart:convert';
|
|
import 'dart:io';
|
|
import 'package:flutter/foundation.dart';
|
|
import 'package:flutter_sound/flutter_sound.dart';
|
|
import 'package:http/http.dart' as http;
|
|
import 'package:path_provider/path_provider.dart';
|
|
import 'package:permission_handler/permission_handler.dart';
|
|
import 'package:mymuseum_visitapp/Services/Glasses/engines/stt_engine.dart';
|
|
|
|
/// STT via API Whisper — format OpenAI multipart/form-data.
|
|
///
|
|
/// Compatible avec :
|
|
/// - OpenAI Whisper API : endpoint = 'https://api.openai.com/v1/audio/transcriptions'
|
|
/// - faster-whisper-server : endpoint = 'http://ton-serveur:8000/v1/audio/transcriptions'
|
|
/// - openedai-speech : même format
|
|
///
|
|
/// Coût : $0.006/min (OpenAI) ou gratuit si auto-hébergé.
|
|
class WhisperSttEngine implements SttEngine {
|
|
final String endpoint;
|
|
final String apiKey;
|
|
final String model;
|
|
|
|
final FlutterSoundRecorder _recorder = FlutterSoundRecorder();
|
|
bool _initialized = false;
|
|
String? _currentPath;
|
|
|
|
WhisperSttEngine({
|
|
required this.apiKey,
|
|
this.endpoint = 'https://api.openai.com/v1/audio/transcriptions',
|
|
this.model = 'whisper-1',
|
|
});
|
|
|
|
Future<void> _ensureInitialized() async {
|
|
if (_initialized) return;
|
|
await Permission.microphone.request();
|
|
await _recorder.openRecorder();
|
|
_initialized = true;
|
|
}
|
|
|
|
@override
|
|
Future<String> transcribeOnce({
|
|
required String languageCode,
|
|
Duration timeout = const Duration(seconds: 8),
|
|
}) async {
|
|
await _ensureInitialized();
|
|
|
|
final dir = await getTemporaryDirectory();
|
|
_currentPath = '${dir.path}/whisper_stt_${DateTime.now().millisecondsSinceEpoch}.wav';
|
|
|
|
try {
|
|
await _recorder.startRecorder(
|
|
toFile: _currentPath,
|
|
codec: Codec.pcm16WAV,
|
|
sampleRate: 16000,
|
|
numChannels: 1,
|
|
);
|
|
|
|
await Future.delayed(timeout);
|
|
await _recorder.stopRecorder();
|
|
|
|
return await _transcribe(_currentPath!, languageCode);
|
|
} catch (e) {
|
|
debugPrint('[WhisperSttEngine] error: $e');
|
|
try { await _recorder.stopRecorder(); } catch (_) {}
|
|
return '';
|
|
} finally {
|
|
try {
|
|
if (_currentPath != null) File(_currentPath!).deleteSync();
|
|
} catch (_) {}
|
|
_currentPath = null;
|
|
}
|
|
}
|
|
|
|
@override
|
|
Future<void> cancel() async {
|
|
try { await _recorder.stopRecorder(); } catch (_) {}
|
|
}
|
|
|
|
Future<String> _transcribe(String wavPath, String languageCode) async {
|
|
final lang = languageCode.split('-').first; // "fr-FR" → "fr"
|
|
final bytes = await File(wavPath).readAsBytes();
|
|
|
|
final request = http.MultipartRequest('POST', Uri.parse(endpoint))
|
|
..headers['Authorization'] = 'Bearer $apiKey'
|
|
..fields['model'] = model
|
|
..fields['language'] = lang
|
|
..fields['response_format'] = 'json'
|
|
..files.add(http.MultipartFile.fromBytes(
|
|
'file',
|
|
bytes,
|
|
filename: 'audio.wav',
|
|
));
|
|
|
|
final response = await request.send().timeout(const Duration(seconds: 15));
|
|
final body = await response.stream.bytesToString();
|
|
|
|
if (response.statusCode != 200) {
|
|
debugPrint('[WhisperSttEngine] HTTP ${response.statusCode}: $body');
|
|
return '';
|
|
}
|
|
|
|
final json = jsonDecode(body) as Map<String, dynamic>;
|
|
final text = (json['text'] as String? ?? '').trim();
|
|
debugPrint('[WhisperSttEngine] "$text"');
|
|
return text;
|
|
}
|
|
}
|