aws_transcribe_streaming 0.1.0

AWS Transcribe Streaming client for producing real-time transcriptions for your media content using HTTP/2.

AWS Transcribe Streaming client for producing real-time transcriptions for your media content using HTTP/2.

Getting started

Add necessary dependencies to pubspec.yaml:

dependencies:
  aws_common: ^0.6.0
  aws_transcribe_streaming: ^0.1.0

copied to clipboard

Obtain a pair of Access/Secret keys for the AWS IAM user with transcribe:StartStreamTranscription permission. See details in AWS documentation: Transcribing streaming audio

It is recommended to use Temporary security credentials with session token obtained from the backend just before starting the transcribing process.

See also best practices to improve streaming transcription efficiency.

Usage

1. Create a new transcribe streaming client:

import 'package:aws_common/aws_common.dart';
import 'package:aws_transcribe_streaming/aws_transcribe_streaming.dart';

final transcribeStreamingClient = TranscribeStreamingClient(
  region: 'eu-central-1',
  credentialsProvider: StaticCredentialsProvider(AWSCredentials(
    'ASIAIOEXAMPLEEXAMPLE',                       // accessKeyId
    'wJalrXUtnFEMI/K7MDENG/bPxRfiCYEXAMPLEKEY',   // secretAccessKey
    'AQoDYXdzEJr...',                             // sessionToken
    DateTime.now().add(const Duration(hours: 1)), // expiration
  )),
);

copied to clipboard

2. Start a stream transcription:

final (response, audioStreamSink, transcriptEventStream) =
    await transcribeStreamingClient.startStreamTranscription(
  const StartStreamTranscriptionRequest(
    languageCode: LanguageCode.enUs,
    mediaSampleRateHertz: 48000,
    mediaEncoding: MediaEncoding.pcm,
  ),
);

copied to clipboard

3. Subscribe to a raw TranscriptEvent stream:

final transcriptSubscription = transcriptEventStream
    .listen((TranscriptEvent event) => print(event));

copied to clipboard

or use a custom strategy to decode TranscriptEvents and build the realtime transcription:

final transcriptSubscription = transcriptEventStream
    .transform(const TranscriptEventStreamDecoder(PlainTextTranscriptionStrategy()))
    .listen((String message) => print(message));

copied to clipboard

4. Start sending audio data to the audio stream sink:

// Raw audio data from the microphone in PCM signed 16-bit little-endian audio format
Stream<Uint8List> audioStream;

final audioStreamSubscription = audioStream.listen(
  audioStreamSink.add,
  onDone: audioStreamSink.close,
);

copied to clipboard