byte[] 可以转换成连续的输入流

Question

由 Netty 构建的项目 Websocket 服务器
Netty 客户端发送请求：

File file = new File("D:\\zh-16000-30s.pcm");
FileInputStream fis = new FileInputStream(file);

int length = 0;
int dataSize = 4096;
byte[] bytes = new byte[dataSize];

int status = 0;
// simulator Andorid or IOS push Streaming 
while ((length = fis.read(bytes, 0, dataSize)) != -1) {

    JSONObject jsonObject = new JSONObject();
    jsonObject.put("audio", Base64.getEncoder().encodeToString(Arrays.copyOf(bytes, length)));\\
    jsonObject.put("status", status);
    WebSocketFrame frame = new TextWebSocketFrame(jsonObject.toJSONString());
    ch.writeAndFlush(frame);
    status = 1;
}

if(length == -1){
    status = 2;
}
if(status == 2){
    JSONObject jsonObject = new JSONObject();
    jsonObject.put("audio", "");
    jsonObject.put("status", status);
    WebSocketFrame frame = new TextWebSocketFrame(jsonObject.toJSONString());
    ch.writeAndFlush(frame);
}

Netty 服务器处理程序：

protected void channelRead0(ChannelHandlerContext ctx, WebSocketFrame frame) throws Exception {
        // ping and pong frames already handled

        if (frame instanceof TextWebSocketFrame) {
            // Send the uppercase string back.
            String request = ((TextWebSocketFrame) frame).text();
            JSONObject jsonObject = JSONObject.parseObject(request);
            Integer status = jsonObject.getInteger("status");
            byte[] recByte = Base64.getDecoder().decode(jsonObject.getString("audio"));
            if(status.intValue() == 0){
                ctx.channel().attr(AttributeKey.newInstance("login")).getAndSet(recByte);
            }else if(status.intValue() == 1){
                byte[] a = (byte[]) ctx.channel().attr(AttributeKey.valueOf("login")).get();
                byte[] c=new byte[a.length+recByte.length];  
                System.arraycopy(a, 0, c, 0, a.length);  
                System.arraycopy(recByte, 0, c, a.length, recByte.length); 
                ctx.channel().attr(AttributeKey.valueOf("login")).getAndSet(c);
            }else if(status.intValue() == 2){
                // the end of file or streaming 
                saveAudio((byte[]) ctx.channel().attr(AttributeKey.valueOf("login")).get());

            }

            ctx.channel().writeAndFlush(new TextWebSocketFrame(request.toUpperCase(Locale.US)));
        } else {
            String message = "unsupported frame type: " + frame.getClass().getName();
            throw new UnsupportedOperationException(message);
        }
    }

我想使用 Microsoft Speech Streaming Recognition

示例代码sn-p：

// Creates an instance of a speech config with specified
    // subscription key and service region. Replace with your own subscription key
    // and service region (e.g., "westus").
    SpeechConfig config = SpeechConfig.fromSubscription("YourSubscriptionKey", "YourServiceRegion");

    // Create an audio stream from a wav file.
    // Replace with your own audio file name.
    PullAudioInputStreamCallback callback = new **WavStream**(new FileInputStream("YourAudioFile.wav"));
    AudioConfig audioInput = AudioConfig.fromStreamInput(callback);

代码 sn-p 2：

private final InputStream stream;

public WavStream(InputStream wavStream) {
    try {
        this.stream = parseWavHeader(wavStream);
    } catch (Exception ex) {
        throw new IllegalArgumentException(ex.getMessage());
    }
}

@Override
public int read(byte[] dataBuffer) {
    long ret = 0;

    try {
        ret = this.stream.read(dataBuffer, 0, dataBuffer.length);
    } catch (Exception ex) {
        System.out.println("Read " + ex);
    }

    return (int)Math.max(0, ret);
}

@Override
public void close() {
    try {
        this.stream.close();
    } catch (IOException ex) {
        // ignored
    }
}

问题：

如何将 byte[] 转换为 inputStream 连续。

例如：

1. 我说30s的声音，假设1s等于netty服务器接收一次包

2. netty服务器发送1s包到微软语音识别

3. 微软语音服务器返回中间结果

4. netty 客户端发送完成，微软同时识别

谢谢

Answer 1

你的问题是关于 netty websocket 服务器的吗？或者关于语音 SDK 对象？

我对以这种方式使用语音 SDK 的建议是使用推送流而不是拉流。通常，您更容易管理。伪代码：

// FOR SETUP... BEFORE you are accepting audio in your websocket server
//              (or on first acceptance of the first packet of audio):
// create push stream
// create audio config from push stream
// create speech config
// create speech recognizer from speech config and audio config
// hook up event handlers for intermediate results (recognizing events)
// hook up event handlers for final results (recognized events)
// start recognition (recognize once or start continuous recognition)

// ON EACH AUDIO packet your websocket server accepts:
// push the audio data into the push stream with

// ON EACH recognizing event, send back the result.text to your client
// ON EACH recognized event, send back the result.text to your client

--抢劫室 [MSFT]