Exam AWS Certified Machine Learning - Specialty topic 1 question 215 discussion

https://docs.aws.amazon.com/transcribe/latest/dg/improving-accuracy.html

A: Only specific words can be corrected

AE is most straight forward.

In contrast, Amazon Transcribe with custom vocabularies (option A) and Amazon Translate (option E) provide a simpler, more efficient solution with lower operational overhead, making them better suited for the company’s needs.

from AWS docs. No doubt. B is a solution that can fix more problem than this requirement but no need.

It is clear from AWS docs. No doubt.

custom language model will be needed as custom vocab will just help with pronunciation, and requirement clearly states handling domain specific terminologies, which cannot be handled by custom vocab.

Option A - Amazon Transcribe with custom vocabularies, allows you to enhance the transcription accuracy by providing domain-specific terminology, names, and locations. Custom vocabularies enable you to train the transcription model to recognize specific words or phrases commonly used in the context of sports commentary. This would help ensure that the transcriptions accurately capture the specialized terminology and context of the commentary, meeting the requirements of the sports broadcasting company. Additionally, the option mentions supporting options to provide tuning data, which further enhances the flexibility and customization of the solution.

Could someone explain why A is not correct? As par AWS documentation: Use custom vocabularies to improve transcription accuracy for one or more specific words. These are generally domain-specific terms, such as brand names and acronyms, proper nouns, and words that Amazon Transcribe isn't rendering correctly. Custom vocabularies can be used with all supported languages.

A - Yes - Transcribe custom vocabs allow domain specific transforms B - No - Custom language models are typically used for fine-tuning for specific accents, dialects, or unique speech patterns. This question is about domain specific terminology C - No - building and tarining requires more overhead D - No - building and tarining requires more overhead E - Once transcribed in english, translate can perform laguage transformation

A. NO - Amazon Transcribe with custom vocabularies does not allow to take into account the broader context (https://docs.aws.amazon.com/transcribe/latest/dg/custom-vocabulary-create-list.html) B. YES - Custom language models are designed to improve transcription accuracy for domain-specific speech (https://docs.aws.amazon.com/transcribe/latest/dg/custom-language-models.html) C. NO - better to use built-in Translate service than base Seq2Seq D. NO - Hugging Face Speech2Text is custom model, use standard Transcribe E. YES - we need to translate English

custom language + Translate

Answer BE

BE : For Specific Language: Custom Language Model

Comentary context is custom language

It's BE, custom languages is for domain-specific speech like terminologies, custom vocabulary is for words liken nouns.

Two sub-processes are needed: Speech to Text and Text to Text. We can consider Amazon Transcribe for Speech2Text. If we use custom language models or SageMaker, we would need to gather our own data to train or retrain models. for less effort, (a) option is better than (b) option. Then, Amazon Translate can be used to translate transcription to other language: (e) option