Gere uma transcrição de um arquivo usando um SDK AWS

PDF

Os exemplos de código a seguir mostram como gerar uma transcrição de um arquivo de áudio de origem usando o streaming do Amazon Transcribe.

C++

SDK para C++

nota

Tem mais sobre GitHub. Encontre o exemplo completo e saiba como configurar e executar no Repositório de exemplos de código da AWS.

int main() {
    Aws::SDKOptions options;

    Aws::InitAPI(options);
    {
        //TODO(User): Set to the region of your AWS account.
        const Aws::String region = Aws::Region::US_WEST_2;

        //Load a profile that has been granted AmazonTranscribeFullAccess AWS managed permission policy.
        Aws::Client::ClientConfiguration config;
#ifdef _WIN32
        // ATTENTION: On Windows with the AWS C++ SDK, this example only runs if the SDK is built
        // with the curl library. 
        // For more information, see the accompanying ReadMe.
        // For more information, see "Building the SDK for Windows with curl".
        // https://docs.aws.amazon.com/sdk-for-cpp/v1/developer-guide/setup-windows.html
        //TODO(User): Update to the location of your .crt file.
        config.caFile = "C:/curl/bin/curl-ca-bundle.crt";
#endif
        config.region = region;

        TranscribeStreamingServiceClient client(config);
        StartStreamTranscriptionHandler handler;
        handler.SetOnErrorCallback(
                [](const Aws::Client::AWSError<TranscribeStreamingServiceErrors> &error) {
                        std::cerr << "ERROR: " + error.GetMessage() << std::endl;
                });
        //SetTranscriptEventCallback called for every 'chunk' of file transcripted.
        // Partial results are returned in real time.
        handler.SetTranscriptEventCallback([](const TranscriptEvent &ev) {
                for (auto &&r: ev.GetTranscript().GetResults()) {
                    if (r.GetIsPartial()) {
                        std::cout << "[partial] ";
                    }
                    else {
                        std::cout << "[Final] ";
                    }
                    for (auto &&alt: r.GetAlternatives()) {
                        std::cout << alt.GetTranscript() << std::endl;
                    }
                }
        });

        StartStreamTranscriptionRequest request;
        request.SetMediaSampleRateHertz(SAMPLE_RATE);
        request.SetLanguageCode(LanguageCode::en_US);
        request.SetMediaEncoding(
                MediaEncoding::pcm); // wav and aiff files are PCM formats.
        request.SetEventStreamHandler(handler);

        auto OnStreamReady = [](AudioStream &stream) {
                Aws::FStream file(FILE_NAME, std::ios_base::in | std::ios_base::binary);
                if (!file.is_open()) {
                    std::cerr << "Failed to open " << FILE_NAME << '\n';
                }
                std::array<char, BUFFER_SIZE> buf;
                int i = 0;
                while (file) {
                    file.read(&buf[0], buf.size());

                    if (!file)
                        std::cout << "File: only " << file.gcount() << " could be read"
                                  << std::endl;

                    Aws::Vector<unsigned char> bits{buf.begin(), buf.end()};
                    AudioEvent event(std::move(bits));
                    if (!stream) {
                        std::cerr << "Failed to create a stream" << std::endl;
                        break;
                    }
                    //The std::basic_istream::gcount() is used to count the characters in the given string. It returns
                    //the number of characters extracted by the last read() operation.
                    if (file.gcount() > 0) {
                        if (!stream.WriteAudioEvent(event)) {
                            std::cerr << "Failed to write an audio event" << std::endl;
                            break;
                        }
                    }
                    else {
                        break;
                    }
                    std::this_thread::sleep_for(std::chrono::milliseconds(
                            25)); // Slow down because we are streaming from a file.
                }
                if (!stream.WriteAudioEvent(
                        AudioEvent())) {
                    // Per the spec, we have to send an empty event (an event without a payload) at the end.
                    std::cerr << "Failed to send an empty frame" << std::endl;
                }
                else {
                    std::cout << "Successfully sent the empty frame" << std::endl;
                }
                stream.flush();
                stream.Close();
        };

        Aws::Utils::Threading::Semaphore signaling(0 /*initialCount*/, 1 /*maxCount*/);
        auto OnResponseCallback = [&signaling](
                const TranscribeStreamingServiceClient * /*unused*/,
                const Model::StartStreamTranscriptionRequest & /*unused*/,
                const Model::StartStreamTranscriptionOutcome &outcome,
                const std::shared_ptr<const Aws::Client::AsyncCallerContext> & /*unused*/) {

                if (!outcome.IsSuccess()) {
                    std::cerr << "Transcribe streaming error "
                              << outcome.GetError().GetMessage() << std::endl;
                }

                signaling.Release();
        };

        std::cout << "Starting..." << std::endl;
        client.StartStreamTranscriptionAsync(request, OnStreamReady, OnResponseCallback,
                                             nullptr /*context*/);
        signaling.WaitOne(); // Prevent the application from exiting until we're done.
        std::cout << "Done" << std::endl;
    }

    Aws::ShutdownAPI(options);

    return 0;
}

• Para obter detalhes da API, consulte StartStreamTranscriptiona Referência AWS SDK for C++ da API.

Java

SDK para Java 2.x

nota

Tem mais sobre GitHub. Encontre o exemplo completo e saiba como configurar e executar no Repositório de exemplos de código da AWS.

/**
 * To run this AWS code example, ensure that you have set up your development
 * environment, including your AWS credentials.
 *
 * For information, see this documentation topic:
 *
 * https://docs.aws.amazon.com/sdk-for-java/latest/developer-guide/get-started.html
 */

public class TranscribeStreamingDemoFile {
    private static final Region REGION = Region.US_EAST_1;
    private static TranscribeStreamingAsyncClient client;

    public static void main(String args[]) throws ExecutionException, InterruptedException {

        final String USAGE = "\n" +
                "Usage:\n" +
                "    <file> \n\n" +
                "Where:\n" +
                "    file - the location of a PCM file to transcribe. In this example, ensure the PCM file is 16 hertz (Hz). \n";

        if (args.length != 1) {
            System.out.println(USAGE);
            System.exit(1);
        }

        String file = args[0];
        client = TranscribeStreamingAsyncClient.builder()
                .region(REGION)
                .build();

        CompletableFuture<Void> result = client.startStreamTranscription(getRequest(16_000),
                new AudioStreamPublisher(getStreamFromFile(file)),
                getResponseHandler());

        result.get();
        client.close();
    }

    private static InputStream getStreamFromFile(String file) {
        try {
            File inputFile = new File(file);
            InputStream audioStream = new FileInputStream(inputFile);
            return audioStream;

        } catch (FileNotFoundException e) {
            throw new RuntimeException(e);
        }
    }

    private static StartStreamTranscriptionRequest getRequest(Integer mediaSampleRateHertz) {
        return StartStreamTranscriptionRequest.builder()
                .languageCode(LanguageCode.EN_US)
                .mediaEncoding(MediaEncoding.PCM)
                .mediaSampleRateHertz(mediaSampleRateHertz)
                .build();
    }

    private static StartStreamTranscriptionResponseHandler getResponseHandler() {
        return StartStreamTranscriptionResponseHandler.builder()
                .onResponse(r -> {
                    System.out.println("Received Initial response");
                })
                .onError(e -> {
                    System.out.println(e.getMessage());
                    StringWriter sw = new StringWriter();
                    e.printStackTrace(new PrintWriter(sw));
                    System.out.println("Error Occurred: " + sw.toString());
                })
                .onComplete(() -> {
                    System.out.println("=== All records stream successfully ===");
                })
                .subscriber(event -> {
                    List<Result> results = ((TranscriptEvent) event).transcript().results();
                    if (results.size() > 0) {
                        if (!results.get(0).alternatives().get(0).transcript().isEmpty()) {
                            System.out.println(results.get(0).alternatives().get(0).transcript());
                        }
                    }
                })
                .build();
    }

    private static class AudioStreamPublisher implements Publisher<AudioStream> {
        private final InputStream inputStream;
        private static Subscription currentSubscription;

        private AudioStreamPublisher(InputStream inputStream) {
            this.inputStream = inputStream;
        }

        @Override
        public void subscribe(Subscriber<? super AudioStream> s) {

            if (this.currentSubscription == null) {
                this.currentSubscription = new SubscriptionImpl(s, inputStream);
            } else {
                this.currentSubscription.cancel();
                this.currentSubscription = new SubscriptionImpl(s, inputStream);
            }
            s.onSubscribe(currentSubscription);
        }
    }

    public static class SubscriptionImpl implements Subscription {
        private static final int CHUNK_SIZE_IN_BYTES = 1024 * 1;
        private final Subscriber<? super AudioStream> subscriber;
        private final InputStream inputStream;
        private ExecutorService executor = Executors.newFixedThreadPool(1);
        private AtomicLong demand = new AtomicLong(0);

        SubscriptionImpl(Subscriber<? super AudioStream> s, InputStream inputStream) {
            this.subscriber = s;
            this.inputStream = inputStream;
        }

        @Override
        public void request(long n) {
            if (n <= 0) {
                subscriber.onError(new IllegalArgumentException("Demand must be positive"));
            }

            demand.getAndAdd(n);

            executor.submit(() -> {
                try {
                    do {
                        ByteBuffer audioBuffer = getNextEvent();
                        if (audioBuffer.remaining() > 0) {
                            AudioEvent audioEvent = audioEventFromBuffer(audioBuffer);
                            subscriber.onNext(audioEvent);
                        } else {
                            subscriber.onComplete();
                            break;
                        }
                    } while (demand.decrementAndGet() > 0);
                } catch (Exception e) {
                    subscriber.onError(e);
                }
            });
        }

        @Override
        public void cancel() {
            executor.shutdown();
        }

        private ByteBuffer getNextEvent() {
            ByteBuffer audioBuffer = null;
            byte[] audioBytes = new byte[CHUNK_SIZE_IN_BYTES];

            int len = 0;
            try {
                len = inputStream.read(audioBytes);

                if (len <= 0) {
                    audioBuffer = ByteBuffer.allocate(0);
                } else {
                    audioBuffer = ByteBuffer.wrap(audioBytes, 0, len);
                }
            } catch (IOException e) {
                throw new UncheckedIOException(e);
            }

            return audioBuffer;
        }

        private AudioEvent audioEventFromBuffer(ByteBuffer bb) {
            return AudioEvent.builder()
                    .audioChunk(SdkBytes.fromByteBuffer(bb))
                    .build();
        }
    }
}

• Para obter detalhes da API, consulte StartStreamTranscriptiona Referência AWS SDK for Java 2.x da API.

Swift

SDK para Swift

nota

Tem mais sobre GitHub. Encontre o exemplo completo e saiba como configurar e executar no Repositório de exemplos de código da AWS.

Use o streaming do Amazon Transcribe para transcrever o idioma falado em um arquivo de áudio.

/// An example that demonstrates how to watch an transcribe event stream to
/// transcribe audio from a file to the console.

import ArgumentParser
import AWSClientRuntime
import AWSTranscribeStreaming
import Foundation

/// Identify one of the media file formats supported by Amazon Transcribe.
enum TranscribeFormat: String, ExpressibleByArgument {
    case ogg = "ogg"
    case pcm = "pcm"
    case flac = "flac"
}

// -MARK: - Async command line tool

struct ExampleCommand: ParsableCommand {
    // -MARK: Command arguments
    @Flag(help: "Show partial results")
    var showPartial = false
    @Option(help: "Language code to transcribe into")
    var lang: String = "en-US"
    @Option(help: "Format of the source audio file")
    var format: TranscribeFormat
    @Option(help: "Sample rate of the source audio file in Hertz")
    var sampleRate: Int = 16000
    @Option(help: "Path of the source audio file")
    var path: String
    @Option(help: "Name of the Amazon S3 Region to use (default: us-east-1)")
    var region = "us-east-1"

    static var configuration = CommandConfiguration(
        commandName: "tsevents",
        abstract: """
        This example shows how to use event streaming with Amazon Transcribe.
        """,
        discussion: """
        """
    )

    /// Create and return an Amazon Transcribe audio stream from the file
    /// specified in the arguments.
    /// 
    /// - Throws: Errors from `TranscribeError`.
    ///
    /// - Returns: `AsyncThrowingStream<TranscribeStreamingClientTypes.AudioStream, Error>`
    func createAudioStream() async throws
                -> AsyncThrowingStream<TranscribeStreamingClientTypes.AudioStream, Error> {

        let fileURL: URL = URL(fileURLWithPath: path)
        let audioData = try Data(contentsOf: fileURL)

        // Properties defining the size of audio chunks and the total size of
        // the audio file in bytes. You should try to send chunks that last on
        // average 125 milliseconds.

        let chunkSizeInMilliseconds = 125.0
        let chunkSize = Int(chunkSizeInMilliseconds  / 1000.0 * Double(sampleRate) * 2.0)
        let audioDataSize = audioData.count

        // Create an audio stream from the source data. The stream's job is
        // to send the audio in chunks to Amazon Transcribe as
        // `AudioStream.audioevent` events.

        let audioStream = AsyncThrowingStream<TranscribeStreamingClientTypes.AudioStream,
                                Error> { continuation in
            Task {
                var currentStart = 0
                var currentEnd = min(chunkSize, audioDataSize - currentStart)

                // Generate and send chunks of audio data as `audioevent`
                // events until the entire file has been sent. Each event is
                // yielded to the SDK after being created.

                while currentStart < audioDataSize {
                    let dataChunk = audioData[currentStart ..< currentEnd]
                    
                    let audioEvent = TranscribeStreamingClientTypes.AudioStream.audioevent(
                        .init(audioChunk: dataChunk)
                    )
                    let yieldResult = continuation.yield(audioEvent)
                    switch yieldResult {
                        case .enqueued(_):
                            // The chunk was successfully enqueued into the
                            // stream. The `remaining` parameter estimates how
                            // much room is left in the queue, but is ignored here.
                            break
                        case .dropped(_):
                            // The chunk was dropped because the queue buffer
                            // is full. This will cause transcription errors.
                            print("Warning: Dropped audio! The transcription will be incomplete.")
                        case .terminated:
                            print("Audio stream terminated.")
                            continuation.finish()
                            return
                        default:
                            print("Warning: Unrecognized response during audio streaming.")
                    }

                    currentStart = currentEnd
                    currentEnd = min(currentStart + chunkSize, audioDataSize)
                }

                // Let the SDK's continuation block know the stream is over.

                continuation.finish()
            }
        }

        return audioStream
    }

    /// Run the transcription process.
    ///
    /// - Throws: An error from `TranscribeError`.
    func transcribe(encoding: TranscribeStreamingClientTypes.MediaEncoding) async throws {
        // Create the Transcribe Streaming client.

        let client = TranscribeStreamingClient(
            config: try await TranscribeStreamingClient.TranscribeStreamingClientConfiguration(
                region: region
            )
        )

        // Start the transcription running on the audio stream.

        let output = try await client.startStreamTranscription(
            input: StartStreamTranscriptionInput(
                audioStream: try await createAudioStream(),
                languageCode: TranscribeStreamingClientTypes.LanguageCode(rawValue: lang),
                mediaEncoding: encoding,
                mediaSampleRateHertz: sampleRate
            )
        )

        // Iterate over the events in the returned transcript result stream.
        // Each `transcriptevent` contains a list of result fragments which
        // need to be concatenated together to build the final transcript.
        for try await event in output.transcriptResultStream! {
            switch event {
            case .transcriptevent(let event):
            for result in event.transcript?.results ?? [] {
                guard let transcript = result.alternatives?.first?.transcript else {
                    continue
                }

                // If showing partial results is enabled and the result is
                // partial, show it. Partial results may be incomplete, and
                // may be inaccurate, with upcoming audio making the
                // transcription complete or by giving more context to make
                // transcription make more sense.

                if (result.isPartial && showPartial) {
                    print("[Partial] \(transcript)")
                }

                // When the complete fragment of transcribed text is ready,
                // print it. This could just as easily be used to draw the
                // text as a subtitle over a playing video, though timing
                // would need to be managed.

                if !result.isPartial {
                    if (showPartial) {
                        print("[Final  ] ", terminator: "")
                    }
                    print(transcript)
                }
            }
            default:
                print("Error: Unexpected message from Amazon Transcribe:")
            }
        }
    }

    /// Convert the value of the `--format` command line option into the
    /// corresponding Transcribe Streaming `MediaEncoding` type.
    ///
    /// - Returns: The `MediaEncoding` equivalent of the format specified on
    ///   the command line.
    func getMediaEncoding() -> TranscribeStreamingClientTypes.MediaEncoding {
        let mediaEncoding: TranscribeStreamingClientTypes.MediaEncoding
        
        switch format {
        case .flac:
            mediaEncoding = .flac
        case .ogg:
            mediaEncoding = .oggOpus
        case .pcm:
            mediaEncoding = .pcm
        }

        return mediaEncoding
    }
}

// -MARK: - Entry point

/// The program's asynchronous entry point.
@main
struct Main {
    static func main() async {
        let args = Array(CommandLine.arguments.dropFirst())

        do {
            let command = try ExampleCommand.parse(args)
            try await command.transcribe(encoding: command.getMediaEncoding())
        } catch let error as TranscribeError {
            print("ERROR: \(error.errorDescription ?? "Unknown error")")
        } catch {
            ExampleCommand.exit(withError: error)
        }
    }    
}

/// Errors thrown by the example's functions.
enum TranscribeError: Error {
    /// No transcription stream available.
    case noTranscriptionStream
    /// The source media file couldn't be read.
    case readError

    var errorDescription: String? {
        switch self {
        case .noTranscriptionStream:
            return "No transcription stream returned by Amazon Transcribe."
        case .readError:
            return "Unable to read the source audio file."
        }
    }
}

• Para obter detalhes da API, consulte StartStreamTranscriptiona referência da API AWS SDK for Swift.