Google文字转语音延迟

Question

I am creating a real-time voice application that involves the Google Text-To-Speech service. However, I am getting latencies of between 600-1100ms which is far too slow for my application. The audio is only around 3 seconds long, how can I improven this? (That latency is a measure of how long it take for me to send the request and then receive the audio).

UPDATE

The code I am using is:

//I call this at the start of my program
TTSclient = TextToSpeechClient.Create();

//This is the method that I call everytime I make a TTS call in my program
public static Google.Protobuf.ByteString MakeTTS(string text)
    {
        SynthesisInput input = new SynthesisInput
        {
            Text = text
        };
        VoiceSelectionParams voice = new VoiceSelectionParams
        {
            LanguageCode = "en-AU",
            Name = "en-AU-Wavenet-A"
        };
        AudioConfig config = new AudioConfig
        {
            AudioEncoding = AudioEncoding.Linear16,
            SampleRateHertz = 16000,
            SpeakingRate = 0.9
        };
        var TTSresponse = TTSclient.SynthesizeSpeech(new SynthesizeSpeechRequest
        {
            Input = input,
            Voice = voice,
            AudioConfig = config
        });
        return TTSresponse.AudioContent;
    }

Thanks

Answer 1

I recommend to check first the Latency median by API method in the metrics page of the TTS API . If you see there that the latency is between 600 to 1,100 ms, then I don't see much to do because all requests are done synchronously and since this is a shared resource, the SLA for those APIs only covers availability, not latency.

If the results you get there are way lower, then I can only think of two things that may slow down your results: the network's own latency or any additional processing being done. If the latest is the case, then you would have to try and error different settings for your request (for example, I would wonder if specifying a device profile, since this feature is currently in beta, would probably result in a slightly slower response).