Azure Text to Speech调用Riff格式时如何静默获取音频数据？

使用Azure认知服务的文本转语音服务，通过SSML构建请求并调用SpeakSsmlAsync函数时，选择Audio24Khz160KBitRateMonoMp3输出格式会直接返回语音数据；但选择Riff24Khz16BitMonoPcm格式时，函数会先通过扬声器播放语音，再返回数据。需要实现静默调用该Riff格式，仅获取语音数据不播放。

2024年8月3日更新，相关代码如下：

相关代码 #

// 初始化SpeechConfig
SpeechConfig speechConfig = SpeechConfig.FromSubscription(SubscriptionKey, SubscriptionRegion);
speechConfig.OutputFormat = OutputFormat.Detailed;
speechConfig.SetSpeechSynthesisOutputFormat(SpeechSynthesisOutputFormat.Riff24Khz16BitMonoPcm);
speechConfig.SpeechSynthesisVoiceName = "de-DE-KatjaNeural";
speechConfig.SetProperty(PropertyId.Speech_LogFilename, LogServices.SpeechLogFilepath);

// 创建SpeechSynthesizer并调用服务
using (SpeechSynthesizer speechSynthesizer = new SpeechSynthesizer(speechConfig))
{
    // 生成SSML文本
    string strSsml = _textToSSML(Language, speechConfig.SpeechSynthesisVoiceName, strText);

    // 调用文本转语音接口
    SpeechSynthesisResult speechSynthesisResult = await speechSynthesizer.SpeakSsmlAsync(strSsml);

    // 处理返回结果
    if (speechSynthesisResult.Reason == ResultReason.SynthesizingAudioCompleted)
    {
        // 将WAV数据处理并保存为MP3
        WaveFile waveFile = new WaveFile(speechSynthesisResult.AudioData);
        _processAndSave(waveFile, audioFilepaths);
    }
    else
    {
        // 记录错误日志
        new LogServices().AddTextToSpeechError(speechSynthesisResult, strText);
    }
}

解决方案 #

问题根源在于SpeechSynthesizer的默认行为：仅传入SpeechConfig实例时，SDK会自动调用系统默认音频输出设备播放合成语音。要实现静默获取数据，只需在创建SpeechSynthesizer时传入null作为AudioConfig参数，明确禁用音频播放功能。

修改后的核心代码如下：

// 创建SpeechSynthesizer时传入null禁用音频播放
using (SpeechSynthesizer speechSynthesizer = new SpeechSynthesizer(speechConfig, null))
{
    // 后续逻辑保持不变
    string strSsml = _textToSSML(Language, speechConfig.SpeechSynthesisVoiceName, strText);
    SpeechSynthesisResult speechSynthesisResult = await speechSynthesizer.SpeakSsmlAsync(strSsml);
    
    if (speechSynthesisResult.Reason == ResultReason.SynthesizingAudioCompleted)
    {
        WaveFile waveFile = new WaveFile(speechSynthesisResult.AudioData);
        _processAndSave(waveFile, audioFilepaths);
    }
    else
    {
        new LogServices().AddTextToSpeechError(speechSynthesisResult, strText);
    }
}

修改后，无论选择哪种音频输出格式，SpeakSsmlAsync都会直接返回语音数据，不会触发扬声器播放。

内容的提问来源于stack exchange，提问作者Richard