Amazon Transcribe

Overview

Automatically converts speech to text
Uses a deep learning process called Automatic Speech Recognition (ASR) to convert speech to text quickly and accurately
Accepts audio input in FLAC, MP3, MP4, or WAV formats
Supports batch transcription and real-time streaming (HTTP/2, WebSockets)
Streaming audio is supported for English, French, and Spanish
Can automatically detect the dominant language spoken in the audio
Can identify and distinguish speakers (speaker diarization)
Supports channel identification
- Example: two callers on a call can be transcribed separately
Supports PII Redaction
- Automatically removes personally identifiable information such as names, age, or social security numbers

Custom Vocabulary (for words)
- Helps Transcribe recognize domain-specific or non-standard terms
- Useful for technical words, acronyms, jargon, etc.
- Two types:
  - Vocabulary Lists: simple list of special words
  - Vocabulary Tables: advanced control using SoundsLike, IPA and DisplayAs
- Example:
  - Without custom vocabulary: “AWS Microservices” → “USA my crow services”
  - With custom vocabulary: “AWS Microservices” → correct transcription
Custom Language Models (for context)
- Provide context, not new words
- Train the Transcribe model using your own domain-specific text data
- Helps Transcribe choose the correct meaning based on industry or use case
- Example:
  - “microservice”:
    - Bird context → “my crow service”
    - IT context → “microservice”
Using both custom vocabularies and custom language models provides the highest transcription accuracy

Amazon Transcribe includes a machine learning-powered toxicity detection feature
Detects toxicity using two types of cues:
- Speech cues: tone, pitch, and emotion (e.g., angry or aggressive tone)
- Text-based cues: transcription analysis for profanities, hate speech, etc.
Toxicity Categories: Sexual harassment, Hate speech, Threats, Abuse, Profanity, Insults, Graphic content

Automatically converts medical-related speech to text (HIPAA compliant)
Has the ability to transcribe medical terminologies such as:
- Medicine names
- Procedures
- Conditions and diseases
Supports both real-time (microphone) and batch (upload multiple files) transcriptions
Automatically converts medical-related speech to text (HIPAA compliant)
Specialized version trained on medical terminology
Has ability to transcribe medical terminologies such as:
- Medicine names
- Procedures
- Conditions and diseases
Supports:
- Real-time transcription (microphone)
- Batch transcription (uploaded audio files)

Call Analytics:
- Trained specifically for customer service and sales calls
- Real-time transcriptions and insights
- Sentiment, talk speed, interruptions, look for specific phrases
Medical:
- Trained on medical terminology
- HIPAA-eligible
Subtitling & Media
- Live subtitle output
- Generates metadata for media assets
- Enables searchable media archives