Aliyun 语音识别&语音合成
导入 SDK
将ZIP包中的nuisdk.framework添加到工程中，并在工程Build Phases的Link Binary With Libraries中添加nuisdk.framework。请确保在编译配置的General > Frameworks, Libraries, and Embedded Content中配置nuisdk.framework为Embed & Sign。
调用步骤
- 初始化SDK、录音实例。设置代理,初始化录音机
// 初始化语音转文字
instance?.nui_initialize(initParams.utf8String, logLevel: LOG_LEVEL_ERROR, saveLog: true)
instance?.nui_set_params(sttParams.utf8String)
instance?.delegate = self
voiceRecorder = NlsVoiceRecorder()
voiceRecorder?.delegate = self
- 根据业务需求配置参数。初始化参数合成和识别都是一样的。
/// SDK初始化参数
internal var initParams: NSString {
let bundle = Bundle.main.path(forResource: "Resources", ofType: "bundle")!
let bundlePath = Bundle(path: bundle)!.resourcePath
let idString = ASIdentifierManager.shared().advertisingIdentifier.uuidString
voicePath = debugPath()
var dict: [String: String] = [:]
dict["workspace"] = bundlePath // 必填
dict["debug_path"] = voicePath
dict["device_id"] = idString // 必填
dict["save_wav"] = "true"
// 从阿里云获取appkey和token进行语音服务访问
dict["app_key"] = appkey
if token.isEmpty {
HGToast("Need token")
return ""
}
dict["token"] = token
dict["url"] = "wss://nls-gateway.cn-shanghai.aliyuncs.com:443/ws/v1"
// FullMix = 0 // 选用此模式开启本地功能并需要进行鉴权注册
// FullCloud = 1 // 在线实时语音识别可以选这个
// FullLocal = 2 // 选用此模式开启本地功能并需要进行鉴权注册
// AsrMix = 3 // 选用此模式开启本地功能并需要进行鉴权注册
// AsrCloud = 4 // 在线一句话识别可以选这个
// AsrLocal = 5 // 选用此模式开启本地功能并需要进行鉴权注册
dict["service_mode"] = "1" // 必填
var jsonStr = ""
do {
let data = try JSONSerialization.data(withJSONObject: dict, options: .prettyPrinted)
jsonStr = String(data: data, encoding: .utf8) ?? ""
} catch {
print("error genInitParams:\(error.localizedDescription)")
}
return jsonStr as NSString
}
- 调用nui_dialog_start开始识别。
instance?.nui_dialog_start(MODE_P2T, dialogParam: NSString(string: "").utf8String)
- 根据音频状态回调audio_state_changed_callback，打开录音机。
func onNuiAudioStateChanged(_ state: NuiAudioState) {
print("onNuiAudioStateChanged state=\(state.rawValue)")
if state == STATE_CLOSE || state == STATE_PAUSE {
voiceRecorder?.stop(true)
} else if state == STATE_OPEN {
recordedVoiceData = NSMutableData()
voiceRecorder?.start()
}
}
- 在user_data_callback回调中提供录音数据。
-(int)onNuiNeedAudioData:(char *)audioData length:(int)len data:(NSMutableData *) recordedVoiceData{
static int emptyCount = 0;
@autoreleasepool {
@synchronized(recordedVoiceData){
if (recordedVoiceData.length > 0) {
int recorder_len = 0;
if (recordedVoiceData.length > len)
recorder_len = len;
else
recorder_len = recordedVoiceData.length;
NSData *tempData = [recordedVoiceData subdataWithRange:NSMakeRange(0, recorder_len)];
[tempData getBytes:audioData length:recorder_len];
tempData = nil;
NSInteger remainLength = recordedVoiceData.length - recorder_len;
NSRange range = NSMakeRange(recorder_len, remainLength);
[recordedVoiceData setData:[recordedVoiceData subdataWithRange:range]];
emptyCount = 0;
return recorder_len;
} else {
if (emptyCount++ >= 50) {
emptyCount = 0;
}
return 0;
}
}
}
return 0;
}
- 在EVENT_ASR_PARTIAL_RESULT和EVENT_SENTENCE_END事件回调中获取识别结果。并将结果转为需要的 String。
func onNuiEventCallback(_ nuiEvent: NuiCallbackEvent, dialog: Int, kwsResult wuw: UnsafePointer<CChar>!, asrResult asr_result: UnsafePointer<CChar>!, ifFinish finish: Bool, retCode code: Int32) {
print("onNuiEventCallback event: \(nuiEvent) finish: \(finish)")
var result = NSString()
if nuiEvent == EVENT_ASR_PARTIAL_RESULT || nuiEvent == EVENT_ASR_RESULT {
result = NSString(utf8String: asr_result) ?? NSString()
print("RESULT: \(result) finish: \(finish)")
if let jsonData = String(result).data(using: .utf8) {
do {
let jsonObject = try JSONSerialization.jsonObject(with: jsonData, options: [])
if let jsonDict = jsonObject as? [String: Any] {
if let payload = jsonDict["payload"] as? [String: Any] {
if let res = payload["result"] as? String {
DispatchQueue.main.async {
self.onDone?(res)
}
}
}
}
} catch {
print("Error parsing JSON: \(error.localizedDescription)")
}
}
} else if nuiEvent == EVENT_ASR_ERROR {
print("EVENT_ASR_ERROR error: \(code)")
return
} else if nuiEvent == EVENT_MIC_ERROR {
print("MIC ERROR")
voiceRecorder?.stop(true)
voiceRecorder?.start()
return
}
// finish 为真（可能是发生错误，也可能是完成识别）表示一次任务生命周期结束，可以开始新的识别
if finish {
print("STT result: \(String(result))")
}
}
- 调用nui_dialog_cancel结束识别。
instance?.nui_dialog_cancel(false)
- 结束调用，使用nui_release接口释放SDK资源。
// 释放资源
deinit {
instance?.nui_release()
tts?.nui_tts_release()
}
TTS 步骤与之类似
主要是获取 data 的方法是：
// 需要合成数据
func onNuiTtsUserdataCallback(_ info: UnsafeMutablePointer<CChar>!, infoLen info_len: Int32, buffer: UnsafeMutablePointer<CChar>!, len: Int32, taskId task_id: UnsafeMutablePointer<CChar>!) {
if info_len > 0 {
print("onNuiTtsUserdataCallback info text \(String(describing: info)). index: \(info_len)")
}
if len > 0 {
voicePlayer.write(buffer, length: len)
print("哈哈哈哈哈:\(len)")
}
}
然后通过voicePlayer播放出来