Skip to content

Incorrect Transcription in cases which include the digit "7"

0

I’m using AWS Transcribe Service to convert user audio input into text. It works perfectly in almost all cases, except when the spoken input contains numbers, particularly those that include the digit “7.”

Whenever a number ends with “7,” the transcription result is incorrect. For example: Saying “6759372” gets transcribed as “67593772.” Saying “6759373” gets transcribed as “67593773.”

Here’s how I’m using the service:

self.transcribe_client.start_stream_transcription( language_code="en-IN", media_sample_rate_hz=8000, media_encoding="pcm", vocabulary_name=None, session_id=str(uuid.uuid4()), enable_partial_results_stabilization=True, partial_results_stability="high" )

I’ve tried adjusting these parameters, but the issue still persists.

1 Answer
0

Based on the information provided, I don't see any specific documentation about issues with the digit "7" in Amazon Transcribe. The transcription behavior you're experiencing where "7" is being duplicated in number sequences is not a documented limitation or known issue in the AWS Transcribe service.

Amazon Transcribe does have specific handling for numbers, which varies by language. For languages with number transcription support, Amazon Transcribe converts spoken numbers into their digit form rather than word form. Amazon Transcribe Medical, for example, has specific rules for transcribing digits as numbers instead of words, such as converting cardinal numbers greater than 10 to numbers.

The issue you're experiencing might be related to:

  1. The specific language model for "en-IN" (English-India) and how it processes certain number pronunciations
  2. Audio quality or clarity issues when pronouncing the digit "7"
  3. Potential configuration issues with the streaming transcription parameters

I would recommend:

  • Testing with different language codes to see if the issue persists
  • Checking if the issue occurs in batch transcription jobs as well as streaming
  • Ensuring your audio input has sufficient clarity and appropriate volume levels
  • Reviewing the failure reasons that Amazon Transcribe provides if jobs are failing

If the issue continues, you may want to contact AWS Support as this appears to be a specific behavior not documented in the service information.
Sources
Transcribing numbers and punctuation - Amazon Transcribe
Transcribing numbers - Amazon Transcribe
TranscriptionJob - Amazon Transcribe

answered 5 months ago

You are not logged in. Log in to post an answer.

A good answer clearly answers the question and provides constructive feedback and encourages professional growth in the question asker.