Transcrever áudio da entrada de streaming

Nesta seção, demonstramos como transcrever áudio de streaming, como a entrada de um microfone, para texto.

O reconhecimento de voz por streaming permite que você faça streaming de áudio para Speech-to-Text e receba uma transmissão de reconhecimento de voz em tempo real conforme o áudio é processado. Consulte também os limites de áudio para transmissão de solicitações de reconhecimento de fala em streaming. O reconhecimento de fala em streaming só está disponível via gRPC.

Como realizar reconhecimento de fala em streaming em um arquivo local

Veja abaixo um exemplo de reconhecimento de fala em streaming em um arquivo de áudio local. Há um limite de 10 MB em todas as solicitações de streaming enviadas à API. Esse limite se aplica à solicitação StreamingRecognize inicial e ao tamanho de cada mensagem individual no stream. Exceder esse limite causará um erro.

Go

Para aprender a instalar e usar a biblioteca de cliente da Speech-to-Text, consulte Bibliotecas de cliente da Speech-to-Text. Para mais informações, consulte a documentação de referência da API Speech-to-Text Go.

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

import (
	"context"
	"flag"
	"fmt"
	"io"
	"log"
	"os"
	"path/filepath"

	speech "cloud.google.com/go/speech/apiv1"
	"cloud.google.com/go/speech/apiv1/speechpb"
)

func main() {
	flag.Usage = func() {
		fmt.Fprintf(os.Stderr, "Usage: %s <AUDIOFILE>\n", filepath.Base(os.Args[0]))
		fmt.Fprintf(os.Stderr, "<AUDIOFILE> must be a path to a local audio file. Audio file must be a 16-bit signed little-endian encoded with a sample rate of 16000.\n")

	}
	flag.Parse()
	if len(flag.Args()) != 1 {
		log.Fatal("Please pass path to your local audio file as a command line argument")
	}
	audioFile := flag.Arg(0)

	ctx := context.Background()

	client, err := speech.NewClient(ctx)
	if err != nil {
		log.Fatal(err)
	}
	stream, err := client.StreamingRecognize(ctx)
	if err != nil {
		log.Fatal(err)
	}
	// Send the initial configuration message.
	if err := stream.Send(&speechpb.StreamingRecognizeRequest{
		StreamingRequest: &speechpb.StreamingRecognizeRequest_StreamingConfig{
			StreamingConfig: &speechpb.StreamingRecognitionConfig{
				Config: &speechpb.RecognitionConfig{
					Encoding:        speechpb.RecognitionConfig_LINEAR16,
					SampleRateHertz: 16000,
					LanguageCode:    "en-US",
				},
			},
		},
	}); err != nil {
		log.Fatal(err)
	}

	f, err := os.Open(audioFile)
	if err != nil {
		log.Fatal(err)
	}
	defer f.Close()

	go func() {
		buf := make([]byte, 1024)
		for {
			n, err := f.Read(buf)
			if n > 0 {
				if err := stream.Send(&speechpb.StreamingRecognizeRequest{
					StreamingRequest: &speechpb.StreamingRecognizeRequest_AudioContent{
						AudioContent: buf[:n],
					},
				}); err != nil {
					log.Printf("Could not send audio: %v", err)
				}
			}
			if err == io.EOF {
				// Nothing else to pipe, close the stream.
				if err := stream.CloseSend(); err != nil {
					log.Fatalf("Could not close stream: %v", err)
				}
				return
			}
			if err != nil {
				log.Printf("Could not read from %s: %v", audioFile, err)
				continue
			}
		}
	}()

	for {
		resp, err := stream.Recv()
		if err == io.EOF {
			break
		}
		if err != nil {
			log.Fatalf("Cannot stream results: %v", err)
		}
		if err := resp.Error; err != nil {
			log.Fatalf("Could not recognize: %v", err)
		}
		for _, result := range resp.Results {
			fmt.Printf("Result: %+v\n", result)
		}
	}
}

Java

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

/**
 * Performs streaming speech recognition on raw PCM audio data.
 *
 * @param fileName the path to a PCM audio file to transcribe.
 */
public static void streamingRecognizeFile(String fileName) throws Exception, IOException {
  Path path = Paths.get(fileName);
  byte[] data = Files.readAllBytes(path);

  // Instantiates a client with GOOGLE_APPLICATION_CREDENTIALS
  try (SpeechClient speech = SpeechClient.create()) {

    // Configure request with local raw PCM audio
    RecognitionConfig recConfig =
        RecognitionConfig.newBuilder()
            .setEncoding(AudioEncoding.LINEAR16)
            .setLanguageCode("en-US")
            .setSampleRateHertz(16000)
            .setModel("default")
            .build();
    StreamingRecognitionConfig config =
        StreamingRecognitionConfig.newBuilder().setConfig(recConfig).build();

    class ResponseApiStreamingObserver<T> implements ApiStreamObserver<T> {
      private final SettableFuture<List<T>> future = SettableFuture.create();
      private final List<T> messages = new java.util.ArrayList<T>();

      @Override
      public void onNext(T message) {
        messages.add(message);
      }

      @Override
      public void onError(Throwable t) {
        future.setException(t);
      }

      @Override
      public void onCompleted() {
        future.set(messages);
      }

      // Returns the SettableFuture object to get received messages / exceptions.
      public SettableFuture<List<T>> future() {
        return future;
      }
    }

    ResponseApiStreamingObserver<StreamingRecognizeResponse> responseObserver =
        new ResponseApiStreamingObserver<>();

    BidiStreamingCallable<StreamingRecognizeRequest, StreamingRecognizeResponse> callable =
        speech.streamingRecognizeCallable();

    ApiStreamObserver<StreamingRecognizeRequest> requestObserver =
        callable.bidiStreamingCall(responseObserver);

    // The first request must **only** contain the audio configuration:
    requestObserver.onNext(
        StreamingRecognizeRequest.newBuilder().setStreamingConfig(config).build());

    // Subsequent requests must **only** contain the audio data.
    requestObserver.onNext(
        StreamingRecognizeRequest.newBuilder()
            .setAudioContent(ByteString.copyFrom(data))
            .build());

    // Mark transmission as completed after sending the data.
    requestObserver.onCompleted();

    List<StreamingRecognizeResponse> responses = responseObserver.future().get();

    for (StreamingRecognizeResponse response : responses) {
      // For streaming recognize, the results list has one is_final result (if available) followed
      // by a number of in-progress results (if iterim_results is true) for subsequent utterances.
      // Just print the first result here.
      StreamingRecognitionResult result = response.getResultsList().get(0);
      // There can be several alternative transcripts for a given chunk of speech. Just use the
      // first (most likely) one here.
      SpeechRecognitionAlternative alternative = result.getAlternativesList().get(0);
      System.out.printf("Transcript : %s\n", alternative.getTranscript());
    }
  }
}

Node.js

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

const fs = require('fs');

// Imports the Google Cloud client library
const speech = require('@google-cloud/speech');

// Creates a client
const client = new speech.SpeechClient();

/**
 * TODO(developer): Uncomment the following lines before running the sample.
 */
// const filename = 'Local path to audio file, e.g. /path/to/audio.raw';
// const encoding = 'Encoding of the audio file, e.g. LINEAR16';
// const sampleRateHertz = 16000;
// const languageCode = 'BCP-47 language code, e.g. en-US';

const request = {
  config: {
    encoding: encoding,
    sampleRateHertz: sampleRateHertz,
    languageCode: languageCode,
  },
  interimResults: false, // If you want interim results, set this to true
};

// Stream the audio to the Google Cloud Speech API
const recognizeStream = client
  .streamingRecognize(request)
  .on('error', console.error)
  .on('data', data => {
    console.log(
      `Transcription: ${data.results[0].alternatives[0].transcript}`
    );
  });

// Stream an audio file from disk to the Speech API, e.g. "./resources/audio.raw"
fs.createReadStream(filename).pipe(recognizeStream);

Python

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

def transcribe_streaming(stream_file: str) -> speech.RecognitionConfig:
    """Streams transcription of the given audio file using Google Cloud Speech-to-Text API.
    Args:
        stream_file (str): Path to the local audio file to be transcribed.
            Example: "resources/audio.raw"
    """
    client = speech.SpeechClient()

    with open(stream_file, "rb") as audio_file:
        audio_content = audio_file.read()

    # In practice, stream should be a generator yielding chunks of audio data.
    stream = [audio_content]

    requests = (
        speech.StreamingRecognizeRequest(audio_content=chunk) for chunk in stream
    )

    config = speech.RecognitionConfig(
        encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
        sample_rate_hertz=16000,
        language_code="en-US",
    )

    streaming_config = speech.StreamingRecognitionConfig(config=config)

    # streaming_recognize returns a generator.
    responses = client.streaming_recognize(
        config=streaming_config,
        requests=requests,
    )

    for response in responses:
        # Once the transcription has settled, the first result will contain the
        # is_final result. The other results will be for subsequent portions of
        # the audio.
        for result in response.results:
            print(f"Finished: {result.is_final}")
            print(f"Stability: {result.stability}")
            alternatives = result.alternatives
            # The alternatives are ordered from most likely to least.
            for alternative in alternatives:
                print(f"Confidence: {alternative.confidence}")
                print(f"Transcript: {alternative.transcript}")

Outras linguagens

C#: Siga as instruções de configuração do C# na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência do Speech-to-Text para .NET.

PHP: Siga as instruções de configuração do PHP na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência da Speech-to-Text para PHP.

Ruby: Siga as instruções de configuração do Ruby na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência do Speech-to-Text para Ruby.

Ainda que seja possível fazer streaming de um arquivo de áudio local para a API Speech-to-Text, recomendamos realizar o reconhecimento de áudio síncrono ou assíncrono para receber os resultados no modo de lote.

Como realizar reconhecimento de fala em streaming em um stream de áudio

A Speech-to-Text também pode realizar reconhecimento em streaming, áudio em tempo real.

Este é um exemplo de reconhecimento de fala em streaming em um stream de áudio recebido de um microfone:

Go

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

import (
	"context"
	"fmt"
	"io"
	"log"
	"os"

	speech "cloud.google.com/go/speech/apiv1"
	"cloud.google.com/go/speech/apiv1/speechpb"
)

func main() {
	ctx := context.Background()

	client, err := speech.NewClient(ctx)
	if err != nil {
		log.Fatal(err)
	}
	stream, err := client.StreamingRecognize(ctx)
	if err != nil {
		log.Fatal(err)
	}
	// Send the initial configuration message.
	if err := stream.Send(&speechpb.StreamingRecognizeRequest{
		StreamingRequest: &speechpb.StreamingRecognizeRequest_StreamingConfig{
			StreamingConfig: &speechpb.StreamingRecognitionConfig{
				Config: &speechpb.RecognitionConfig{
					Encoding:        speechpb.RecognitionConfig_LINEAR16,
					SampleRateHertz: 16000,
					LanguageCode:    "en-US",
				},
			},
		},
	}); err != nil {
		log.Fatal(err)
	}

	go func() {
		// Pipe stdin to the API.
		buf := make([]byte, 1024)
		for {
			n, err := os.Stdin.Read(buf)
			if n > 0 {
				if err := stream.Send(&speechpb.StreamingRecognizeRequest{
					StreamingRequest: &speechpb.StreamingRecognizeRequest_AudioContent{
						AudioContent: buf[:n],
					},
				}); err != nil {
					log.Printf("Could not send audio: %v", err)
				}
			}
			if err == io.EOF {
				// Nothing else to pipe, close the stream.
				if err := stream.CloseSend(); err != nil {
					log.Fatalf("Could not close stream: %v", err)
				}
				return
			}
			if err != nil {
				log.Printf("Could not read from stdin: %v", err)
				continue
			}
		}
	}()

	for {
		resp, err := stream.Recv()
		if err == io.EOF {
			break
		}
		if err != nil {
			log.Fatalf("Cannot stream results: %v", err)
		}
		if err := resp.Error; err != nil {
			// Workaround while the API doesn't give a more informative error.
			if err.Code == 3 || err.Code == 11 {
				log.Print("WARNING: Speech recognition request exceeded limit of 60 seconds.")
			}
			log.Fatalf("Could not recognize: %v", err)
		}
		for _, result := range resp.Results {
			fmt.Printf("Result: %+v\n", result)
		}
	}
}

Python

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.


import queue
import re
import sys

from google.cloud import speech

import pyaudio

# Audio recording parameters
RATE = 16000
CHUNK = int(RATE / 10)  # 100ms


class MicrophoneStream:
    """Opens a recording stream as a generator yielding the audio chunks."""

    def __init__(self: object, rate: int = RATE, chunk: int = CHUNK) -> None:
        """The audio -- and generator -- is guaranteed to be on the main thread."""
        self._rate = rate
        self._chunk = chunk

        # Create a thread-safe buffer of audio data
        self._buff = queue.Queue()
        self.closed = True

    def __enter__(self: object) -> object:
        self._audio_interface = pyaudio.PyAudio()
        self._audio_stream = self._audio_interface.open(
            format=pyaudio.paInt16,
            # The API currently only supports 1-channel (mono) audio
            # https://goo.gl/z757pE
            channels=1,
            rate=self._rate,
            input=True,
            frames_per_buffer=self._chunk,
            # Run the audio stream asynchronously to fill the buffer object.
            # This is necessary so that the input device's buffer doesn't
            # overflow while the calling thread makes network requests, etc.
            stream_callback=self._fill_buffer,
        )

        self.closed = False

        return self

    def __exit__(
        self: object,
        type: object,
        value: object,
        traceback: object,
    ) -> None:
        """Closes the stream, regardless of whether the connection was lost or not."""
        self._audio_stream.stop_stream()
        self._audio_stream.close()
        self.closed = True
        # Signal the generator to terminate so that the client's
        # streaming_recognize method will not block the process termination.
        self._buff.put(None)
        self._audio_interface.terminate()

    def _fill_buffer(
        self: object,
        in_data: object,
        frame_count: int,
        time_info: object,
        status_flags: object,
    ) -> object:
        """Continuously collect data from the audio stream, into the buffer.

        Args:
            in_data: The audio data as a bytes object
            frame_count: The number of frames captured
            time_info: The time information
            status_flags: The status flags

        Returns:
            The audio data as a bytes object
        """
        self._buff.put(in_data)
        return None, pyaudio.paContinue

    def generator(self: object) -> object:
        """Generates audio chunks from the stream of audio data in chunks.

        Args:
            self: The MicrophoneStream object

        Returns:
            A generator that outputs audio chunks.
        """
        while not self.closed:
            # Use a blocking get() to ensure there's at least one chunk of
            # data, and stop iteration if the chunk is None, indicating the
            # end of the audio stream.
            chunk = self._buff.get()
            if chunk is None:
                return
            data = [chunk]

            # Now consume whatever other data's still buffered.
            while True:
                try:
                    chunk = self._buff.get(block=False)
                    if chunk is None:
                        return
                    data.append(chunk)
                except queue.Empty:
                    break

            yield b"".join(data)


def listen_print_loop(responses: object) -> str:
    """Iterates through server responses and prints them.

    The responses passed is a generator that will block until a response
    is provided by the server.

    Each response may contain multiple results, and each result may contain
    multiple alternatives; for details, see https://goo.gl/tjCPAU.  Here we
    print only the transcription for the top alternative of the top result.

    In this case, responses are provided for interim results as well. If the
    response is an interim one, print a line feed at the end of it, to allow
    the next result to overwrite it, until the response is a final one. For the
    final one, print a newline to preserve the finalized transcription.

    Args:
        responses: List of server responses

    Returns:
        The transcribed text.
    """
    num_chars_printed = 0
    for response in responses:
        if not response.results:
            continue

        # The `results` list is consecutive. For streaming, we only care about
        # the first result being considered, since once it's `is_final`, it
        # moves on to considering the next utterance.
        result = response.results[0]
        if not result.alternatives:
            continue

        # Display the transcription of the top alternative.
        transcript = result.alternatives[0].transcript

        # Display interim results, but with a carriage return at the end of the
        # line, so subsequent lines will overwrite them.
        #
        # If the previous result was longer than this one, we need to print
        # some extra spaces to overwrite the previous result
        overwrite_chars = " " * (num_chars_printed - len(transcript))

        if not result.is_final:
            sys.stdout.write(transcript + overwrite_chars + "\r")
            sys.stdout.flush()

            num_chars_printed = len(transcript)

        else:
            print(transcript + overwrite_chars)

            # Exit recognition if any of the transcribed phrases could be
            # one of our keywords.
            if re.search(r"\b(exit|quit)\b", transcript, re.I):
                print("Exiting..")
                break

            num_chars_printed = 0

    return transcript


def main() -> None:
    """Transcribe speech from audio file."""
    # See http://g.co/cloud/speech/docs/languages
    # for a list of supported languages.
    language_code = "en-US"  # a BCP-47 language tag

    client = speech.SpeechClient()
    config = speech.RecognitionConfig(
        encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
        sample_rate_hertz=RATE,
        language_code=language_code,
    )

    streaming_config = speech.StreamingRecognitionConfig(
        config=config, interim_results=True
    )

    with MicrophoneStream(RATE, CHUNK) as stream:
        audio_generator = stream.generator()
        requests = (
            speech.StreamingRecognizeRequest(audio_content=content)
            for content in audio_generator
        )

        responses = client.streaming_recognize(streaming_config, requests)

        # Now, put the transcription responses to use.
        listen_print_loop(responses)


if __name__ == "__main__":
    main()

Java

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.

/** Performs microphone streaming speech recognition with a duration of 1 minute. */
public static void streamingMicRecognize() throws Exception {

  ResponseObserver<StreamingRecognizeResponse> responseObserver = null;
  try (SpeechClient client = SpeechClient.create()) {

    responseObserver =
        new ResponseObserver<StreamingRecognizeResponse>() {
          ArrayList<StreamingRecognizeResponse> responses = new ArrayList<>();

          public void onStart(StreamController controller) {}

          public void onResponse(StreamingRecognizeResponse response) {
            responses.add(response);
          }

          public void onComplete() {
            for (StreamingRecognizeResponse response : responses) {
              StreamingRecognitionResult result = response.getResultsList().get(0);
              SpeechRecognitionAlternative alternative = result.getAlternativesList().get(0);
              System.out.printf("Transcript : %s\n", alternative.getTranscript());
            }
          }

          public void onError(Throwable t) {
            System.out.println(t);
          }
        };

    ClientStream<StreamingRecognizeRequest> clientStream =
        client.streamingRecognizeCallable().splitCall(responseObserver);

    RecognitionConfig recognitionConfig =
        RecognitionConfig.newBuilder()
            .setEncoding(RecognitionConfig.AudioEncoding.LINEAR16)
            .setLanguageCode("en-US")
            .setSampleRateHertz(16000)
            .build();
    StreamingRecognitionConfig streamingRecognitionConfig =
        StreamingRecognitionConfig.newBuilder().setConfig(recognitionConfig).build();

    StreamingRecognizeRequest request =
        StreamingRecognizeRequest.newBuilder()
            .setStreamingConfig(streamingRecognitionConfig)
            .build(); // The first request in a streaming call has to be a config

    clientStream.send(request);
    // SampleRate:16000Hz, SampleSizeInBits: 16, Number of channels: 1, Signed: true,
    // bigEndian: false
    AudioFormat audioFormat = new AudioFormat(16000, 16, 1, true, false);
    DataLine.Info targetInfo =
        new Info(
            TargetDataLine.class,
            audioFormat); // Set the system information to read from the microphone audio stream

    if (!AudioSystem.isLineSupported(targetInfo)) {
      System.out.println("Microphone not supported");
      System.exit(0);
    }
    // Target data line captures the audio stream the microphone produces.
    TargetDataLine targetDataLine = (TargetDataLine) AudioSystem.getLine(targetInfo);
    targetDataLine.open(audioFormat);
    targetDataLine.start();
    System.out.println("Start speaking");
    long startTime = System.currentTimeMillis();
    // Audio Input Stream
    AudioInputStream audio = new AudioInputStream(targetDataLine);
    while (true) {
      long estimatedTime = System.currentTimeMillis() - startTime;
      byte[] data = new byte[6400];
      audio.read(data);
      if (estimatedTime > 60000) { // 60 seconds
        System.out.println("Stop speaking.");
        targetDataLine.stop();
        targetDataLine.close();
        break;
      }
      request =
          StreamingRecognizeRequest.newBuilder()
              .setAudioContent(ByteString.copyFrom(data))
              .build();
      clientStream.send(request);
    }
  } catch (Exception e) {
    System.out.println(e);
  }
  responseObserver.onComplete();
}

Node.js

Para essas amostras, é preciso que você instale o SoX e que ele esteja disponível no $PATH.

Para Mac OS: brew install sox.
Para a maioria das distribuições do Linux: sudo apt-get install sox libsox-fmt-all.
Para Windows: faça o download dos binários.

Para mais informações sobre como instalar e criar um cliente Speech-to-Text, consulte Bibliotecas de cliente Speech-to-Text.

const recorder = require('node-record-lpcm16');

// Imports the Google Cloud client library
const speech = require('@google-cloud/speech');

// Creates a client
const client = new speech.SpeechClient();

/**
 * TODO(developer): Uncomment the following lines before running the sample.
 */
// const encoding = 'Encoding of the audio file, e.g. LINEAR16';
// const sampleRateHertz = 16000;
// const languageCode = 'BCP-47 language code, e.g. en-US';

const request = {
  config: {
    encoding: encoding,
    sampleRateHertz: sampleRateHertz,
    languageCode: languageCode,
  },
  interimResults: false, // If you want interim results, set this to true
};

// Create a recognize stream
const recognizeStream = client
  .streamingRecognize(request)
  .on('error', console.error)
  .on('data', data =>
    process.stdout.write(
      data.results[0] && data.results[0].alternatives[0]
        ? `Transcription: ${data.results[0].alternatives[0].transcript}\n`
        : '\n\nReached transcription time limit, press Ctrl+C\n'
    )
  );

// Start recording and send the microphone input to the Speech API.
// Ensure SoX is installed, see https://www.npmjs.com/package/node-record-lpcm16#dependencies
recorder
  .record({
    sampleRateHertz: sampleRateHertz,
    threshold: 0,
    // Other options, see https://www.npmjs.com/package/node-record-lpcm16#options
    verbose: false,
    recordProgram: 'rec', // Try also "arecord" or "sox"
    silence: '10.0',
  })
  .stream()
  .on('error', console.error)
  .pipe(recognizeStream);

console.log('Listening, press Ctrl+C to stop.');

Outras linguagens

C#: Siga as instruções de configuração do C# na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência do Speech-to-Text para .NET.

PHP: Siga as instruções de configuração do PHP na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência da Speech-to-Text para PHP.

Ruby: Siga as instruções de configuração do Ruby na página das bibliotecas de cliente e, em seguida, acesse a documentação de referência do Speech-to-Text para Ruby.

Executar o reconhecimento de fala em streaming contínuo

Este é um exemplo de reconhecimento de fala em streaming em um stream de áudio contínuo recebido de um microfone:

Python

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.


import queue
import re
import sys
import time

from google.cloud import speech
import pyaudio

# Audio recording parameters
STREAMING_LIMIT = 240000  # 4 minutes
SAMPLE_RATE = 16000
CHUNK_SIZE = int(SAMPLE_RATE / 10)  # 100ms

RED = "\033[0;31m"
GREEN = "\033[0;32m"
YELLOW = "\033[0;33m"


def get_current_time() -> int:
    """Return Current Time in MS.

    Returns:
        int: Current Time in MS.
    """

    return int(round(time.time() * 1000))


class ResumableMicrophoneStream:
    """Opens a recording stream as a generator yielding the audio chunks."""

    def __init__(
        self: object,
        rate: int,
        chunk_size: int,
    ) -> None:
        """Creates a resumable microphone stream.

        Args:
        self: The class instance.
        rate: The audio file's sampling rate.
        chunk_size: The audio file's chunk size.

        returns: None
        """
        self._rate = rate
        self.chunk_size = chunk_size
        self._num_channels = 1
        self._buff = queue.Queue()
        self.closed = True
        self.start_time = get_current_time()
        self.restart_counter = 0
        self.audio_input = []
        self.last_audio_input = []
        self.result_end_time = 0
        self.is_final_end_time = 0
        self.final_request_end_time = 0
        self.bridging_offset = 0
        self.last_transcript_was_final = False
        self.new_stream = True
        self._audio_interface = pyaudio.PyAudio()
        self._audio_stream = self._audio_interface.open(
            format=pyaudio.paInt16,
            channels=self._num_channels,
            rate=self._rate,
            input=True,
            frames_per_buffer=self.chunk_size,
            # Run the audio stream asynchronously to fill the buffer object.
            # This is necessary so that the input device's buffer doesn't
            # overflow while the calling thread makes network requests, etc.
            stream_callback=self._fill_buffer,
        )

    def __enter__(self: object) -> object:
        """Opens the stream.

        Args:
        self: The class instance.

        returns: None
        """
        self.closed = False
        return self

    def __exit__(
        self: object,
        type: object,
        value: object,
        traceback: object,
    ) -> object:
        """Closes the stream and releases resources.

        Args:
        self: The class instance.
        type: The exception type.
        value: The exception value.
        traceback: The exception traceback.

        returns: None
        """
        self._audio_stream.stop_stream()
        self._audio_stream.close()
        self.closed = True
        # Signal the generator to terminate so that the client's
        # streaming_recognize method will not block the process termination.
        self._buff.put(None)
        self._audio_interface.terminate()

    def _fill_buffer(
        self: object,
        in_data: object,
        *args: object,
        **kwargs: object,
    ) -> object:
        """Continuously collect data from the audio stream, into the buffer.

        Args:
        self: The class instance.
        in_data: The audio data as a bytes object.
        args: Additional arguments.
        kwargs: Additional arguments.

        returns: None
        """
        self._buff.put(in_data)
        return None, pyaudio.paContinue

    def generator(self: object) -> object:
        """Stream Audio from microphone to API and to local buffer

        Args:
            self: The class instance.

        returns:
            The data from the audio stream.
        """
        while not self.closed:
            data = []

            if self.new_stream and self.last_audio_input:
                chunk_time = STREAMING_LIMIT / len(self.last_audio_input)

                if chunk_time != 0:
                    if self.bridging_offset < 0:
                        self.bridging_offset = 0

                    if self.bridging_offset > self.final_request_end_time:
                        self.bridging_offset = self.final_request_end_time

                    chunks_from_ms = round(
                        (self.final_request_end_time - self.bridging_offset)
                        / chunk_time
                    )

                    self.bridging_offset = round(
                        (len(self.last_audio_input) - chunks_from_ms) * chunk_time
                    )

                    for i in range(chunks_from_ms, len(self.last_audio_input)):
                        data.append(self.last_audio_input[i])

                self.new_stream = False

            # Use a blocking get() to ensure there's at least one chunk of
            # data, and stop iteration if the chunk is None, indicating the
            # end of the audio stream.
            chunk = self._buff.get()
            self.audio_input.append(chunk)

            if chunk is None:
                return
            data.append(chunk)
            # Now consume whatever other data's still buffered.
            while True:
                try:
                    chunk = self._buff.get(block=False)

                    if chunk is None:
                        return
                    data.append(chunk)
                    self.audio_input.append(chunk)

                except queue.Empty:
                    break

            yield b"".join(data)


def listen_print_loop(responses: object, stream: object) -> None:
    """Iterates through server responses and prints them.

    The responses passed is a generator that will block until a response
    is provided by the server.

    Each response may contain multiple results, and each result may contain
    multiple alternatives; for details, see https://goo.gl/tjCPAU.  Here we
    print only the transcription for the top alternative of the top result.

    In this case, responses are provided for interim results as well. If the
    response is an interim one, print a line feed at the end of it, to allow
    the next result to overwrite it, until the response is a final one. For the
    final one, print a newline to preserve the finalized transcription.

    Arg:
        responses: The responses returned from the API.
        stream: The audio stream to be processed.
    """
    for response in responses:
        if get_current_time() - stream.start_time > STREAMING_LIMIT:
            stream.start_time = get_current_time()
            break

        if not response.results:
            continue

        result = response.results[0]

        if not result.alternatives:
            continue

        transcript = result.alternatives[0].transcript

        result_seconds = 0
        result_micros = 0

        if result.result_end_time.seconds:
            result_seconds = result.result_end_time.seconds

        if result.result_end_time.microseconds:
            result_micros = result.result_end_time.microseconds

        stream.result_end_time = int((result_seconds * 1000) + (result_micros / 1000))

        corrected_time = (
            stream.result_end_time
            - stream.bridging_offset
            + (STREAMING_LIMIT * stream.restart_counter)
        )
        # Display interim results, but with a carriage return at the end of the
        # line, so subsequent lines will overwrite them.

        if result.is_final:
            sys.stdout.write(GREEN)
            sys.stdout.write("\033[K")
            sys.stdout.write(str(corrected_time) + ": " + transcript + "\n")

            stream.is_final_end_time = stream.result_end_time
            stream.last_transcript_was_final = True

            # Exit recognition if any of the transcribed phrases could be
            # one of our keywords.
            if re.search(r"\b(exit|quit)\b", transcript, re.I):
                sys.stdout.write(YELLOW)
                sys.stdout.write("Exiting...\n")
                stream.closed = True
                break
        else:
            sys.stdout.write(RED)
            sys.stdout.write("\033[K")
            sys.stdout.write(str(corrected_time) + ": " + transcript + "\r")

            stream.last_transcript_was_final = False


def main() -> None:
    """start bidirectional streaming from microphone input to speech API"""
    client = speech.SpeechClient()
    config = speech.RecognitionConfig(
        encoding=speech.RecognitionConfig.AudioEncoding.LINEAR16,
        sample_rate_hertz=SAMPLE_RATE,
        language_code="en-US",
        max_alternatives=1,
    )

    streaming_config = speech.StreamingRecognitionConfig(
        config=config, interim_results=True
    )

    mic_manager = ResumableMicrophoneStream(SAMPLE_RATE, CHUNK_SIZE)
    print(mic_manager.chunk_size)
    sys.stdout.write(YELLOW)
    sys.stdout.write('\nListening, say "Quit" or "Exit" to stop.\n\n')
    sys.stdout.write("End (ms)       Transcript Results/Status\n")
    sys.stdout.write("=====================================================\n")

    with mic_manager as stream:
        while not stream.closed:
            sys.stdout.write(YELLOW)
            sys.stdout.write(
                "\n" + str(STREAMING_LIMIT * stream.restart_counter) + ": NEW REQUEST\n"
            )

            stream.audio_input = []
            audio_generator = stream.generator()

            requests = (
                speech.StreamingRecognizeRequest(audio_content=content)
                for content in audio_generator
            )

            responses = client.streaming_recognize(streaming_config, requests)

            # Now, put the transcription responses to use.
            listen_print_loop(responses, stream)

            if stream.result_end_time > 0:
                stream.final_request_end_time = stream.is_final_end_time
            stream.result_end_time = 0
            stream.last_audio_input = []
            stream.last_audio_input = stream.audio_input
            stream.audio_input = []
            stream.restart_counter = stream.restart_counter + 1

            if not stream.last_transcript_was_final:
                sys.stdout.write("\n")
            stream.new_stream = True


if __name__ == "__main__":
    main()

Java

Para autenticar no Speech-to-Text, configure o Application Default Credentials. Para mais informações, consulte Configurar a autenticação para um ambiente de desenvolvimento local.


import com.google.api.gax.rpc.ClientStream;
import com.google.api.gax.rpc.ResponseObserver;
import com.google.api.gax.rpc.StreamController;
import com.google.cloud.speech.v1p1beta1.RecognitionConfig;
import com.google.cloud.speech.v1p1beta1.SpeechClient;
import com.google.cloud.speech.v1p1beta1.SpeechRecognitionAlternative;
import com.google.cloud.speech.v1p1beta1.StreamingRecognitionConfig;
import com.google.cloud.speech.v1p1beta1.StreamingRecognitionResult;
import com.google.cloud.speech.v1p1beta1.StreamingRecognizeRequest;
import com.google.cloud.speech.v1p1beta1.StreamingRecognizeResponse;
import com.google.protobuf.ByteString;
import com.google.protobuf.Duration;
import java.text.DecimalFormat;
import java.util.ArrayList;
import java.util.concurrent.BlockingQueue;
import java.util.concurrent.LinkedBlockingQueue;
import java.util.concurrent.TimeUnit;
import javax.sound.sampled.AudioFormat;
import javax.sound.sampled.AudioSystem;
import javax.sound.sampled.DataLine;
import javax.sound.sampled.DataLine.Info;
import javax.sound.sampled.TargetDataLine;

public class InfiniteStreamRecognize {

  private static final int STREAMING_LIMIT = 290000; // ~5 minutes

  public static final String RED = "\033[0;31m";
  public static final String GREEN = "\033[0;32m";
  public static final String YELLOW = "\033[0;33m";

  // Creating shared object
  private static volatile BlockingQueue<byte[]> sharedQueue = new LinkedBlockingQueue<byte[]>();
  private static TargetDataLine targetDataLine;
  private static int BYTES_PER_BUFFER = 6400; // buffer size in bytes

  private static int restartCounter = 0;
  private static ArrayList<ByteString> audioInput = new ArrayList<ByteString>();
  private static ArrayList<ByteString> lastAudioInput = new ArrayList<ByteString>();
  private static int resultEndTimeInMS = 0;
  private static int isFinalEndTime = 0;
  private static int finalRequestEndTime = 0;
  private static boolean newStream = true;
  private static double bridgingOffset = 0;
  private static boolean lastTranscriptWasFinal = false;
  private static StreamController referenceToStreamController;
  private static ByteString tempByteString;

  public static void main(String... args) {
    InfiniteStreamRecognizeOptions options = InfiniteStreamRecognizeOptions.fromFlags(args);
    if (options == null) {
      // Could not parse.
      System.out.println("Failed to parse options.");
      System.exit(1);
    }

    try {
      infiniteStreamingRecognize(options.langCode);
    } catch (Exception e) {
      System.out.println("Exception caught: " + e);
    }
  }

  public static String convertMillisToDate(double milliSeconds) {
    long millis = (long) milliSeconds;
    DecimalFormat format = new DecimalFormat();
    format.setMinimumIntegerDigits(2);
    return String.format(
        "%s:%s /",
        format.format(TimeUnit.MILLISECONDS.toMinutes(millis)),
        format.format(
            TimeUnit.MILLISECONDS.toSeconds(millis)
                - TimeUnit.MINUTES.toSeconds(TimeUnit.MILLISECONDS.toMinutes(millis))));
  }

  /** Performs infinite streaming speech recognition */
  public static void infiniteStreamingRecognize(String languageCode) throws Exception {

    // Microphone Input buffering
    class MicBuffer implements Runnable {

      @Override
      public void run() {
        System.out.println(YELLOW);
        System.out.println("Start speaking...Press Ctrl-C to stop");
        targetDataLine.start();
        byte[] data = new byte[BYTES_PER_BUFFER];
        while (targetDataLine.isOpen()) {
          try {
            int numBytesRead = targetDataLine.read(data, 0, data.length);
            if ((numBytesRead <= 0) && (targetDataLine.isOpen())) {
              continue;
            }
            sharedQueue.put(data.clone());
          } catch (InterruptedException e) {
            System.out.println("Microphone input buffering interrupted : " + e.getMessage());
          }
        }
      }
    }

    // Creating microphone input buffer thread
    MicBuffer micrunnable = new MicBuffer();
    Thread micThread = new Thread(micrunnable);
    ResponseObserver<StreamingRecognizeResponse> responseObserver = null;
    try (SpeechClient client = SpeechClient.create()) {
      ClientStream<StreamingRecognizeRequest> clientStream;
      responseObserver =
          new ResponseObserver<StreamingRecognizeResponse>() {

            ArrayList<StreamingRecognizeResponse> responses = new ArrayList<>();

            public void onStart(StreamController controller) {
              referenceToStreamController = controller;
            }

            public void onResponse(StreamingRecognizeResponse response) {
              responses.add(response);
              StreamingRecognitionResult result = response.getResultsList().get(0);
              Duration resultEndTime = result.getResultEndTime();
              resultEndTimeInMS =
                  (int)
                      ((resultEndTime.getSeconds() * 1000) + (resultEndTime.getNanos() / 1000000));
              double correctedTime =
                  resultEndTimeInMS - bridgingOffset + (STREAMING_LIMIT * restartCounter);

              SpeechRecognitionAlternative alternative = result.getAlternativesList().get(0);
              if (result.getIsFinal()) {
                System.out.print(GREEN);
                System.out.print("\033[2K\r");
                System.out.printf(
                    "%s: %s [confidence: %.2f]\n",
                    convertMillisToDate(correctedTime),
                    alternative.getTranscript(),
                    alternative.getConfidence());
                isFinalEndTime = resultEndTimeInMS;
                lastTranscriptWasFinal = true;
              } else {
                System.out.print(RED);
                System.out.print("\033[2K\r");
                System.out.printf(
                    "%s: %s", convertMillisToDate(correctedTime), alternative.getTranscript());
                lastTranscriptWasFinal = false;
              }
            }

            public void onComplete() {}

            public void onError(Throwable t) {}
          };
      clientStream = client.streamingRecognizeCallable().splitCall(responseObserver);

      RecognitionConfig recognitionConfig =
          RecognitionConfig.newBuilder()
              .setEncoding(RecognitionConfig.AudioEncoding.LINEAR16)
              .setLanguageCode(languageCode)
              .setSampleRateHertz(16000)
              .build();

      StreamingRecognitionConfig streamingRecognitionConfig =
          StreamingRecognitionConfig.newBuilder()
              .setConfig(recognitionConfig)
              .setInterimResults(true)
              .build();

      StreamingRecognizeRequest request =
          StreamingRecognizeRequest.newBuilder()
              .setStreamingConfig(streamingRecognitionConfig)
              .build(); // The first request in a streaming call has to be a config

      clientStream.send(request);

      try {
        // SampleRate:16000Hz, SampleSizeInBits: 16, Number of channels: 1, Signed: true,
        // bigEndian: false
        AudioFormat audioFormat = new AudioFormat(16000, 16, 1, true, false);
        DataLine.Info targetInfo =
            new Info(
                TargetDataLine.class,
                audioFormat); // Set the system information to read from the microphone audio
        // stream

        if (!AudioSystem.isLineSupported(targetInfo)) {
          System.out.println("Microphone not supported");
          System.exit(0);
        }
        // Target data line captures the audio stream the microphone produces.
        targetDataLine = (TargetDataLine) AudioSystem.getLine(targetInfo);
        targetDataLine.open(audioFormat);
        micThread.start();

        long startTime = System.currentTimeMillis();

        while (true) {

          long estimatedTime = System.currentTimeMillis() - startTime;

          if (estimatedTime >= STREAMING_LIMIT) {

            clientStream.closeSend();
            referenceToStreamController.cancel(); // remove Observer

            if (resultEndTimeInMS > 0) {
              finalRequestEndTime = isFinalEndTime;
            }
            resultEndTimeInMS = 0;

            lastAudioInput = null;
            lastAudioInput = audioInput;
            audioInput = new ArrayList<ByteString>();

            restartCounter++;

            if (!lastTranscriptWasFinal) {
              System.out.print('\n');
            }

            newStream = true;

            clientStream = client.streamingRecognizeCallable().splitCall(responseObserver);

            request =
                StreamingRecognizeRequest.newBuilder()
                    .setStreamingConfig(streamingRecognitionConfig)
                    .build();

            System.out.println(YELLOW);
            System.out.printf("%d: RESTARTING REQUEST\n", restartCounter * STREAMING_LIMIT);

            startTime = System.currentTimeMillis();

          } else {

            if ((newStream) && (lastAudioInput.size() > 0)) {
              // if this is the first audio from a new request
              // calculate amount of unfinalized audio from last request
              // resend the audio to the speech client before incoming audio
              double chunkTime = STREAMING_LIMIT / lastAudioInput.size();
              // ms length of each chunk in previous request audio arrayList
              if (chunkTime != 0) {
                if (bridgingOffset < 0) {
                  // bridging Offset accounts for time of resent audio
                  // calculated from last request
                  bridgingOffset = 0;
                }
                if (bridgingOffset > finalRequestEndTime) {
                  bridgingOffset = finalRequestEndTime;
                }
                int chunksFromMs =
                    (int) Math.floor((finalRequestEndTime - bridgingOffset) / chunkTime);
                // chunks from MS is number of chunks to resend
                bridgingOffset =
                    (int) Math.floor((lastAudioInput.size() - chunksFromMs) * chunkTime);
                // set bridging offset for next request
                for (int i = chunksFromMs; i < lastAudioInput.size(); i++) {
                  request =
                      StreamingRecognizeRequest.newBuilder()
                          .setAudioContent(lastAudioInput.get(i))
                          .build();
                  clientStream.send(request);
                }
              }
              newStream = false;
            }

            tempByteString = ByteString.copyFrom(sharedQueue.take());

            request =
                StreamingRecognizeRequest.newBuilder().setAudioContent(tempByteString).build();

            audioInput.add(tempByteString);
          }

          clientStream.send(request);
        }
      } catch (Exception e) {
        System.out.println(e);
      }
    }
  }
}

Node.js

Para esses exemplos, é preciso instalar o SoX e ele precisa estar disponível no $PATH.

Para Mac OS: brew install sox.
Para a maioria das distribuições do Linux: sudo apt-get install sox libsox-fmt-all.
Para Windows: faça o download dos binários.

Para mais informações sobre como instalar e criar um cliente Speech-to-Text, consulte Bibliotecas de cliente Speech-to-Text.


// const encoding = 'LINEAR16';
// const sampleRateHertz = 16000;
// const languageCode = 'en-US';
// const streamingLimit = 10000; // ms - set to low number for demo purposes

const chalk = require('chalk');
const {Writable} = require('stream');
const recorder = require('node-record-lpcm16');

// Imports the Google Cloud client library
// Currently, only v1p1beta1 contains result-end-time
const speech = require('@google-cloud/speech').v1p1beta1;

const client = new speech.SpeechClient();

const config = {
  encoding: encoding,
  sampleRateHertz: sampleRateHertz,
  languageCode: languageCode,
};

const request = {
  config,
  interimResults: true,
};

let recognizeStream = null;
let restartCounter = 0;
let audioInput = [];
let lastAudioInput = [];
let resultEndTime = 0;
let isFinalEndTime = 0;
let finalRequestEndTime = 0;
let newStream = true;
let bridgingOffset = 0;
let lastTranscriptWasFinal = false;

function startStream() {
  // Clear current audioInput
  audioInput = [];
  // Initiate (Reinitiate) a recognize stream
  recognizeStream = client
    .streamingRecognize(request)
    .on('error', err => {
      if (err.code === 11) {
        // restartStream();
      } else {
        console.error('API request error ' + err);
      }
    })
    .on('data', speechCallback);

  // Restart stream when streamingLimit expires
  setTimeout(restartStream, streamingLimit);
}

const speechCallback = stream => {
  // Convert API result end time from seconds + nanoseconds to milliseconds
  resultEndTime =
    stream.results[0].resultEndTime.seconds * 1000 +
    Math.round(stream.results[0].resultEndTime.nanos / 1000000);

  // Calculate correct time based on offset from audio sent twice
  const correctedTime =
    resultEndTime - bridgingOffset + streamingLimit * restartCounter;

  process.stdout.clearLine();
  process.stdout.cursorTo(0);
  let stdoutText = '';
  if (stream.results[0] && stream.results[0].alternatives[0]) {
    stdoutText =
      correctedTime + ': ' + stream.results[0].alternatives[0].transcript;
  }

  if (stream.results[0].isFinal) {
    process.stdout.write(chalk.green(`${stdoutText}\n`));

    isFinalEndTime = resultEndTime;
    lastTranscriptWasFinal = true;
  } else {
    // Make sure transcript does not exceed console character length
    if (stdoutText.length > process.stdout.columns) {
      stdoutText =
        stdoutText.substring(0, process.stdout.columns - 4) + '...';
    }
    process.stdout.write(chalk.red(`${stdoutText}`));

    lastTranscriptWasFinal = false;
  }
};

const audioInputStreamTransform = new Writable({
  write(chunk, encoding, next) {
    if (newStream && lastAudioInput.length !== 0) {
      // Approximate math to calculate time of chunks
      const chunkTime = streamingLimit / lastAudioInput.length;
      if (chunkTime !== 0) {
        if (bridgingOffset < 0) {
          bridgingOffset = 0;
        }
        if (bridgingOffset > finalRequestEndTime) {
          bridgingOffset = finalRequestEndTime;
        }
        const chunksFromMS = Math.floor(
          (finalRequestEndTime - bridgingOffset) / chunkTime
        );
        bridgingOffset = Math.floor(
          (lastAudioInput.length - chunksFromMS) * chunkTime
        );

        for (let i = chunksFromMS; i < lastAudioInput.length; i++) {
          recognizeStream.write(lastAudioInput[i]);
        }
      }
      newStream = false;
    }

    audioInput.push(chunk);

    if (recognizeStream) {
      recognizeStream.write(chunk);
    }

    next();
  },

  final() {
    if (recognizeStream) {
      recognizeStream.end();
    }
  },
});

function restartStream() {
  if (recognizeStream) {
    recognizeStream.end();
    recognizeStream.removeListener('data', speechCallback);
    recognizeStream = null;
  }
  if (resultEndTime > 0) {
    finalRequestEndTime = isFinalEndTime;
  }
  resultEndTime = 0;

  lastAudioInput = [];
  lastAudioInput = audioInput;

  restartCounter++;

  if (!lastTranscriptWasFinal) {
    process.stdout.write('\n');
  }
  process.stdout.write(
    chalk.yellow(`${streamingLimit * restartCounter}: RESTARTING REQUEST\n`)
  );

  newStream = true;

  startStream();
}
// Start recording and send the microphone input to the Speech API
recorder
  .record({
    sampleRateHertz: sampleRateHertz,
    threshold: 0, // Silence threshold
    silence: 1000,
    keepSilence: true,
    recordProgram: 'rec', // Try also "arecord" or "sox"
  })
  .stream()
  .on('error', err => {
    console.error('Audio recording error ' + err);
  })
  .pipe(audioInputStreamTransform);

console.log('');
console.log('Listening, press Ctrl+C to stop.');
console.log('');
console.log('End (ms)       Transcript Results/Status');
console.log('=========================================================');

startStream();

A seguir

Saiba como avaliar e melhorar a precisão

Faça um teste

Se você ainda não conhece o Google Cloud, crie uma conta para avaliar o desempenho da Speech-to-Text em cenários reais. Clientes novos também recebem US$ 300 em créditos para executar, testar e implantar cargas de trabalho.

Faça um teste gratuito da Speech-to-Text