Skip to content

Latest commit

 

History

History
38 lines (26 loc) · 3.63 KB

File metadata and controls

38 lines (26 loc) · 3.63 KB

Transcription features

The following shows which Amazon Transcribe features are supported - please refer to the service documentation links for further details of how each feature works.

Feature Description Supported
Language identification  Allow Amazon Transcribe to determine the dominant language used in the audio file, which it will then use for the whole transcription process ☑️
Custom vocabularies  Provide more information on how to transcribe specific words or phrases, typically used for domain-specific words for specific use cases ☑️
Vocabulary filtering  Allows you to mark or remove unwanted words from the transcripts ☑️
Custom language models  Use your own text data to improve transcription accuracy for your specific use case
Streaming transcriptions  Send an audio stream to Amazon Transcribe and receive transcription output in real-time
Channel identification  Process each channel in an audio file independently, combining the transcriptions from each channel into a single output ☑️
Speaker diarization  Label each speaker utterance with a distinct speaker tag ☑️
Call analytics  Inject analytical insights from your calls into your transcripts ☑️
Redaction  Mask or remove sensitive personally identifiable information (PII) content from the text transcripts ☑️
KMS-based encryption  Encrypt transcription output files in your Amazon S3 bucket using KMS keys rather than the default Amazon S3 key (SSE-S3)

Feature restrictions

Language support

Some of the Amazon Transcribe features are not available in all supported languages. These are highlighted within the service documentation languages page , and at the time of writing can be summarised as follows:

  • Redaction is only available in the US English en-US language model
  • Call analytics is supported only by a subset of the Amazon Transcribe languages
  • Digit transcription, whereby number phrase like Fifty five or a hundredth are transcribed as 55 and 1/100 respectively, is supported only by a subset of the Amazon Transcribe languages
Custom vocabulary
  • Acronyms are only supported by a subset of the Amazon Transcribe languages
  • Language ID is not currently supported natively with custom vocabularies. However, this solution works around this limitation and performs language ID on clip of the original audio, and then chooses the correct custom vocabulary file before submitting the whole audio file for processing. This workaround will be removed once custom vocabualries are supported when using language ID.
Other feature limitations
  • The Channel identification and Speaker diarization features are mutually exclusive
  • Call analytics requires stereo channel-separated audio files