jm_baidu_stt_plugin

Flutter bindings for the latest Baidu Speech ASR & wake-up SDKs.

• Android: bdasr_V3_20250507_b610f20.jar + native libs (v3.4.5).
• iOS: ASR_iOS_v3.0.12.0.1285b57 static library & resources.

The plugin exposes a simple Dart API that mirrors the reference native demo so you can initialize the SDK, create recognizer/wake-up engines, and handle streaming status callbacks from Dart.

Install

1. Add the dependency to your pubspec.yaml: jm_baidu_stt_plugin: ^0.0.9

2. Install packages: flutter pub get

3. For iOS, run CocoaPods from your host app’s ios/ folder: pod install

iOS SDK delivery (recommended)

Baidu iOS SDK binaries/models are large. For publishing (Git / pub.dev), it is recommended not to commit them into the repository.

This plugin uses CocoaPods prepare_command to download missing artifacts during pod install.

Usage

1. Provide download URLs via environment variables.

Option A (recommended): one zip containing both files

• JM_BAIDU_STT_IOS_SDK_ZIP_URL (the zip must contain libBaiduSpeechSDK.a and bds_easr_input_model.dat somewhere inside)

Option B: two direct file URLs

• JM_BAIDU_STT_IOS_LIB_URL
• JM_BAIDU_STT_IOS_MODEL_URL

2. Run CocoaPods: pod install

If your network requires a proxy, set it before running CocoaPods:

• export https_proxy=http://127.0.0.1:7890
• export http_proxy=http://127.0.0.1:7890

Notes

• If the files already exist in ios/Libs/... and ios/Assets/..., download is skipped (offline friendly).
• To force re-download: export JM_BAIDU_STT_IOS_FORCE_DOWNLOAD=1
• To disable download and fail fast if missing: export JM_BAIDU_STT_IOS_SKIP_DOWNLOAD=1

Tip: set URLs in ios/Podfile (CI/team friendly)

If you don’t want every developer/CI to export env vars manually, you can set them in your host app’s ios/Podfile:

• ENV['JM_BAIDU_STT_IOS_SDK_ZIP_URL'] ||= '<YOUR_ZIP_URL>'

Or (two direct URLs):

• ENV['JM_BAIDU_STT_IOS_LIB_URL'] ||= '<YOUR_LIB_URL>'
• ENV['JM_BAIDU_STT_IOS_MODEL_URL'] ||= '<YOUR_MODEL_URL>'

Quick start

1. Initialize and subscribe to events: JmBaiduSttPlugin.initSDK(...) and provide an onEvent callback.

2. Create the ASR engine: await JmBaiduSttPlugin.create(type: BaiduSpeechBuildType.asr)

3. Start recognition: await JmBaiduSttPlugin.startRecognition()

Optional (some offline modes): await JmBaiduSttPlugin.startRecognition(wakeUpWord: '...')

4. Stop recognition: await JmBaiduSttPlugin.stopRecognition()

5. Wake-up:

• Create: await JmBaiduSttPlugin.create(type: BaiduSpeechBuildType.wakeUp)
• Start: await JmBaiduSttPlugin.startMonitorWakeUp()
• Stop: await JmBaiduSttPlugin.stopMonitorWakeUp()

Event status values

status	meaning
flushData	partial recognition result
statusFinish	final recognition result
volumeChanged	audio level update (0–100)
statusError	recognition/wake-up failure
statusTriggered	wake-up keyword detected

type in the payload is 0 for ASR and 1 for wake-up.

iOS notes (important)

• Simulator: the Baidu static library in this project is arm64 only (device slice). It usually cannot run on iOS Simulator.
• Resource lookup: iOS native code will try to locate .dat files from mainBundle / bundleForClass / common resource bundles, to be resilient under CocoaPods.

iOS ASR defaults

The iOS implementation follows Baidu’s official sample defaults:

• Default productId: 1537
• Default language: EVoiceRecognitionLanguageChinese
• Default strategy: EVR_STRATEGY_ONLINE

If you need parallel mode (online + offline), pass strategy = EVR_STRATEGY_BOTH and ensure the offline model/license files are available.

ASR stability

iOS native layer includes guards for common issues:

• Rapid stopRecognition() → startRecognition() may cause engine is busy: the plugin will auto retry with backoff.
• STOP-induced HTTP timeout will be silently ignored when stopping.
• NoSpeech/Short will be treated as statusFinish with empty string.

Advanced iOS parameters (optional)

The iOS implementation supports extra parameters on create (defaults are applied if not provided):

• productId: defaults to 1537
• language: defaults to EVoiceRecognitionLanguageChinese
• strategy: defaults to EVR_STRATEGY_ONLINE
• enableModelVAD: defaults to false
• offlineEngineType: defaults to EVR_OFFLINE_ENGINE_INPUT (only used when strategy == EVR_STRATEGY_BOTH)
• offlineDatPath: defaults to bundled bds_easr_input_model.dat (only used when strategy == EVR_STRATEGY_BOTH)
• serverUrl: defaults to https://vop.baidu.com/server_api

Current Dart API does not expose these options yet; if you need them you can extend the plugin or invoke MethodChannel('jm_baidu_stt_plugin').invokeMethod('create', ...) yourself.

permission_handler (microphone)

If you use permission_handler on iOS, make sure the microphone permission macro is enabled for the permission_handler_apple target in your host app’s ios/Podfile post_install.

Example lines to add (as recommended by the plugin):

• if target.name == 'permission_handler_apple'
• target.build_configurations.each do |config|
• config.build_settings['GCC_PREPROCESSOR_DEFINITIONS'] ||= ['$(inherited)', 'PERMISSION_MICROPHONE=1']
• end
• end

Android integration notes

• minSdkVersion is set to 24 to align with the new Baidu SDK artifacts.
• The plugin bundles the latest bdasr_V3_20250507_b610f20.jar, WakeUp.bin, and all native .so files (armeabi, armeabi-v7a, arm64-v8a, x86, x86_64).
• Required permissions are already declared in the embedded AndroidManifest.xml, but your host app must still request RECORD_AUDIO at runtime.
• If you shrink/obfuscate, keep the Baidu classes (e.g. add -keep class com.baidu.speech.** { *; } to proguard-rules.pro).

iOS integration notes

• The plugin ships Baidu’s static library (libBaiduSpeechSDK.a), headers, wake-word models, and resource bundles from ASR_iOS_v3.0.12.0.
• Add the following to your app’s Info.plist:
  • NSMicrophoneUsageDescription
  • (Optional) UIBackgroundModes → audio if you record in background.
• Run pod install inside the Flutter app’s ios/ folder after adding the plugin so CocoaPods pulls in the updated podspec.
• The plugin includes a Privacy Manifest (Resources/PrivacyInfo.xcprivacy). Update it if your host app has additional data uses.

Example app

example/lib/main.dart contains a basic control panel showing how to:

1. Initialize the SDK with an event listener.
2. Create ASR and wake-up engines.
3. Start/stop recognition and wake-up monitoring while printing native callback payloads.

Replace the placeholder credentials before running on a device.

License

See LICENSE. The Baidu SDK binaries remain subject to Baidu’s own license terms; ensure you have permission to use them in production.