如何让Android SpeechRecognizer忽略TTS输出，优先识别用户语音？

我使用compileSdk 34、minSdk 33开发，单独使用TextToSpeech（TTS）和SpeechRecognizer时功能正常，但当两者几乎同时启动时，SpeechRecognizer只会识别TTS的输出，完全忽略我在这段时间的语音输入。我需要让识别器专注捕捉用户语音，过滤掉TTS的播放内容。

另外我注意到Android 13及以上机型具备这种能力：比如TTS播放时说出"Hey Google"，我的语音会被识别，TTS自动暂停，Google助手开始监听。想知道如何在自己的APP中实现类似效果。

实验代码 #

package com.example.speechandspeak;

import android.Manifest;
import android.content.Intent;
import android.content.pm.PackageManager;
import android.os.Bundle;
import android.speech.RecognitionListener;
import android.speech.RecognizerIntent;
import android.speech.SpeechRecognizer;
import android.speech.tts.TextToSpeech;
import android.util.Log;
import android.widget.Toast;

import androidx.appcompat.app.AppCompatActivity;
import androidx.core.app.ActivityCompat;
import androidx.core.content.ContextCompat;

import java.util.ArrayList;
import java.util.Locale;

public class MainActivity extends AppCompatActivity implements RecognitionListener {

    private TextToSpeech tts;
    private SpeechRecognizer speechRecognizer;

    @Override
    protected void onCreate(Bundle savedInstanceState) {
        super.onCreate(savedInstanceState);
        setContentView(R.layout.activity_main);

        // Initialize TextToSpeech
        tts = new TextToSpeech(this, new TextToSpeech.OnInitListener() {
            @Override
            public void onInit(int status) {
                if (status == TextToSpeech.SUCCESS) {
                    int ttsLang = tts.setLanguage(Locale.US);
                    if (ttsLang == TextToSpeech.LANG_MISSING_DATA || ttsLang == TextToSpeech.LANG_NOT_SUPPORTED) {
                        Toast.makeText(MainActivity.this, "Language is not supported!", Toast.LENGTH_SHORT).show();
                    } else {
                        speakText();
                    }
                } else {
                    Toast.makeText(MainActivity.this, "TTS Initialization failed!", Toast.LENGTH_SHORT).show();
                }
            }
        });


        // Check if microphone permission is granted
        if (ContextCompat.checkSelfPermission(this, android.Manifest.permission.RECORD_AUDIO) != PackageManager.PERMISSION_GRANTED) {
            // Permission is not granted, request it
            ActivityCompat.requestPermissions(this, new String[]{Manifest.permission.RECORD_AUDIO}, 1);
        }
        // Initialize SpeechRecognizer
        speechRecognizer = SpeechRecognizer.createSpeechRecognizer(this);
        speechRecognizer.setRecognitionListener(this);
    }

    private void speakText() {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
        String message = getString(R.string.tts_message);
        tts.speak(message, TextToSpeech.QUEUE_FLUSH, null, null);

        // After speaking, initiate speech recognition
        Intent speechRecognizerIntent = new Intent(RecognizerIntent.ACTION_RECOGNIZE_SPEECH);
        speechRecognizer.startListening(speechRecognizerIntent);
    }

    @Override
    public void onReadyForSpeech(Bundle params) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onBeginningOfSpeech() {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onRmsChanged(float rmsdB) {
        //Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onBufferReceived(byte[] bytes) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onEndOfSpeech() {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onError(int i) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName() + "(" + i + ")");
    }

    @Override
    public void onEvent(int eventType, Bundle params) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onPartialResults(Bundle partialResults) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    }

    @Override
    public void onResults(Bundle results) {
        Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
        ArrayList<String> matches = results.getStringArrayList(SpeechRecognizer.RESULTS_RECOGNITION);
        if (matches != null && !matches.isEmpty()) {
            Log.v(this.getLocalClassName(),  "Matches: " + matches);
            String text = matches.get(0).toLowerCase(); // Convert to lowercase for case-insensitive comparison
            if (text.equals("stop")) {
                if (tts.isSpeaking()) {
                    tts.stop();
                    speechRecognizer.stopListening(); // Stop listening for further commands
                    Toast.makeText(MainActivity.this, "Stopped!", Toast.LENGTH_SHORT).show();
                }
            }
        }
    }

    @Override
    protected void onDestroy() {
        super.onDestroy();
        if (tts != null) {
            tts.shutdown();
        }
    }
}

Logcat调试信息 #

12:40:09.125 MainActivity              D  Entered: speakText
12:40:09.297                           D  Entered: onReadyForSpeech
12:40:09.761                           D  Entered: onBeginningOfSpeech
12:40:13.365                           D  Entered: onEndOfSpeech
12:40:13.394                           D  Entered: onResults
12:40:13.396                           V  Matches: [demonstration of speech recognition and text to speech say stop to quit]
12:40:14.524 ProfileInstaller          D  Installing profile for com.example.speechandspeak

解决方案 #

1. 配置TTS的AudioAttributes，标记语音合成类型 #

通过设置TTS的音频属性，让系统识别这是语音合成输出，从而在语音识别时自动应用回声消除：

// 在TTS初始化成功后添加以下配置
AudioAttributes audioAttributes = new AudioAttributes.Builder()
        .setUsage(AudioAttributes.USAGE_ASSISTANT) // 标记为助手类语音输出
        .setContentType(AudioAttributes.CONTENT_TYPE_SPEECH) // 内容类型为语音
        .build();
tts.setAudioAttributes(audioAttributes);

2. 给SpeechRecognizer启用回声消除 #

在启动识别的Intent中添加EXTRA_ENABLE_ECHO_CANCELLATION参数，强制启用回声消除功能：

private void speakText() {
    Log.d(this.getLocalClassName(),  "Entered: " + Thread.currentThread().getStackTrace()[2].getMethodName());
    String message = getString(R.string.tts_message);
    tts.speak(message, TextToSpeech.QUEUE_FLUSH, null, null);

    // 修改识别Intent，添加回声消除参数
    Intent speechRecognizerIntent = new Intent(RecognizerIntent.ACTION_RECOGNIZE_SPEECH);
    speechRecognizerIntent.putExtra(RecognizerIntent.EXTRA_LANGUAGE_MODEL, RecognizerIntent.LANGUAGE_MODEL_FREE_FORM);
    speechRecognizerIntent.putExtra(RecognizerIntent.EXTRA_ENABLE_ECHO_CANCELLATION, true); // 启用回声消除
    speechRecognizer.startListening(speechRecognizerIntent);
}

3. 关于Google助手的实现原理 #

Google助手使用的是系统级的VoiceInteractionService，结合硬件级的回声消除技术，并且系统会持续监听唤醒词（如"Hey Google"）。当检测到唤醒词时，系统会通过音频焦点抢占机制暂停正在播放的TTS，然后启动助手的识别流程。第三方APP无法直接使用系统级唤醒词监听，但通过上述两步配置，已经可以让SpeechRecognizer过滤掉自身TTS的输出，专注识别用户语音。

内容的提问来源于stack exchange，提问作者ususer