Prova Gemini 1.5 Pro, il nostro modello multimodale più avanzato di Vertex AI e scopri cosa puoi creare con una finestra di contesto dei token da 1 milione. Prova Gemini 1.5 Pro, il nostro modello multimodale più avanzato di Vertex AI e scopri cosa puoi creare con una finestra di contesto di 1 milione di token.

Trascrizione di audio dall'input in streaming

Questa sezione mostra come trascrivere in testo l'audio in streaming, ad esempio l'input di un microfono.

Il riconoscimento vocale in streaming ti consente di trasmettere audio in streaming a Speech-to-Text e di ricevere i risultati del riconoscimento vocale in streaming in tempo reale durante l'elaborazione dell'audio. Vedi anche i limiti audio per lo streaming delle richieste di riconoscimento vocale. Il riconoscimento vocale in streaming è disponibile soltanto tramite gRPC.

Esecuzione del riconoscimento vocale in streaming su un file locale

Di seguito è riportato un esempio di riconoscimento vocale in streaming su un file audio locale. Esiste un limite di 10 MB su tutte le richieste di streaming inviate all'API. Questo limite si applica sia alla richiesta StreamingRecognize iniziale sia alle dimensioni di ogni singolo messaggio nel flusso. Il superamento di questo limite causerà un errore.

Go

Per scoprire come installare e utilizzare la libreria client per Speech-to-Text, consulta la pagina relativa alle librerie client di Speech-to-Text. Per maggiori informazioni, consulta la documentazione di riferimento dell'API Speech-to-Text Go.

Per eseguire l'autenticazione a Speech-to-Text, configura le Credenziali predefinite dell'applicazione. Per maggiori informazioni, consulta Configurare l'autenticazione per un ambiente di sviluppo locale.

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

/**
 * 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

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

def transcribe_streaming(stream_file: str) -> speech.RecognitionConfig:
    """Streams transcription of the given audio file."""

    client = speech.SpeechClient()

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

    # In practice, stream should be a generator yielding chunks of audio data.
    stream = [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}")

Linguaggi aggiuntivi

C#: segui le istruzioni di configurazione di C# nella pagina delle librerie client e poi consulta la documentazione di riferimento di Speech-to-Text per .NET.

PHP: segui le istruzioni per la configurazione dei file PHP nella pagina delle librerie client e poi consulta la documentazione di riferimento di Speech-to-Text per PHP.

Ruby: segui le istruzioni per la configurazione di Ruby nella pagina delle librerie client e consulta la documentazione di riferimento di Speech-to-Text per Ruby.

Sebbene sia possibile trasmettere in streaming un file audio locale all'API Speech-to-Text, ti consigliamo di eseguire il riconoscimento audio sincrono o asincrone per i risultati della modalità batch.

Esecuzione del riconoscimento vocale in streaming durante uno stream audio

Speech-to-Text può anche eseguire il riconoscimento in streaming audio in tempo reale.

Ecco un esempio di riconoscimento vocale in streaming su uno stream audio ricevuto da un microfono:

Go

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


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

/** 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

Questi esempi richiedono l'installazione di SoX e deve essere disponibile nel tuo $PATH.

Per Mac OS: brew install sox.
Per la maggior parte delle distribuzioni Linux: sudo apt-get install sox libsox-fmt-all.
Per Windows: scarica i file binari.

Per saperne di più sull'installazione e sulla creazione di un client Speech-to-Text, consulta la pagina Librerie client di 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.');

Linguaggi aggiuntivi

C#: segui le istruzioni di configurazione di C# nella pagina delle librerie client e poi consulta la documentazione di riferimento di Speech-to-Text per .NET.

PHP: segui le istruzioni per la configurazione dei file PHP nella pagina delle librerie client e poi consulta la documentazione di riferimento di Speech-to-Text per PHP.

Ruby: segui le istruzioni per la configurazione di Ruby nella pagina delle librerie client e consulta la documentazione di riferimento di Speech-to-Text per Ruby.

Passaggi successivi

Scopri come trascrivere uno stream audio all'infinito

Provalo

Se non hai mai utilizzato Google Cloud, crea un account per valutare le prestazioni di Speech-to-Text in scenari reali. I nuovi clienti ricevono anche 300 $ di crediti gratuiti per l'esecuzione, il test e il deployment dei carichi di lavoro.

Prova Speech-to-Text gratuitamente