AWS Transcribe Engine User Guide

AWS Transcribe provides enterprise-grade, managed speech-to-text capabilities through Kamiwaza's unified transcription API. Audio is processed by AWS's cloud service, eliminating local compute requirements.

Quick Start (UI)

1. In Kamiwaza, go to Models and click Add external endpoint
2. Select AWS Transcribe from the Service dropdown
3. Enter your configuration:
   • Display Name: A name for this transcription endpoint
   • Language: Select a language or use "Auto-detect" (default)
   • AWS Region: Region where audio will be processed
   • S3 Bucket: Bucket for temporary audio storage
   • Credential Secret: AWS credentials (URN or JSON)
4. Click Save Endpoint
5. Deploy the model from the Models list

Prerequisites

1. AWS Account with Transcribe access enabled
2. S3 Bucket for batch transcription (audio is temporarily stored during processing)
3. IAM Credentials with the following permissions:

{
  "Version": "2012-10-17",
  "Statement": [
    {
      "Effect": "Allow",
      "Action": [
        "transcribe:StartTranscriptionJob",
        "transcribe:GetTranscriptionJob",
        "transcribe:DeleteTranscriptionJob"
      ],
      "Resource": "*"
    },
    {
      "Effect": "Allow",
      "Action": ["s3:PutObject", "s3:GetObject", "s3:DeleteObject"],
      "Resource": "arn:aws:s3:::YOUR-BUCKET/transcribe-jobs/*"
    }
  ]
}

Configuration

Create a model with external_endpoint configuration:

{
  "name": "aws-transcribe",
  "capability": "audio_transcription",
  "config": {
    "external_endpoint": {
      "service": "aws_transcribe",
      "region": "us-east-1",
      "language_code": "en-US",
      "s3_bucket": "your-bucket-name",
      "credential_secret_urn": "urn:li:secret:aws-transcribe-creds"
    }
  }
}

Required Fields

Field	Description
region	AWS region (e.g., us-east-1)
language_code	BCP-47 code (e.g., en-US, es-ES, fr-FR)
s3_bucket	S3 bucket for batch audio storage

Optional Fields

Field	Default	Description
media_format	auto	Audio format: auto, mp3, mp4, wav, flac, ogg, amr, webm. "auto" detects from filename.
sample_rate	16000	Sample rate in Hz
show_speaker_labels	false	Enable speaker identification (diarize)
max_speaker_labels	2	Maximum speakers (1-10)
vocabulary_name	-	Custom vocabulary for domain terms
redact_pii	false	Remove PII from transcript
pii_entity_types	[]	PII types to redact: PERSON, EMAIL, PHONE_NUMBER, etc.

Credentials

Provide AWS credentials via one of:

1. Kamiwaza Secret Catalog (recommended):

   First, store your AWS credentials in Kamiwaza's secret catalog, then reference by URN:

   { "credential_secret_urn": "urn:li:secret:aws-transcribe-creds" }

   The secret value should be JSON containing AWS credentials:

   {
     "aws_access_key_id": "AKIA...",
     "aws_secret_access_key": "...",
     "aws_session_token": "..."  // optional, for temporary credentials
   }

2. Inline credentials (for testing):

   Paste the raw JSON directly in the Credential Secret field:

   {"aws_access_key_id":"AKIA...","aws_secret_access_key":"...","aws_session_token":"..."}

3. Environment variables (on the Kamiwaza server):

   • AWS_ACCESS_KEY_ID
   • AWS_SECRET_ACCESS_KEY
   • AWS_SESSION_TOKEN (optional)

Deployment

Deploy the model using engine name aws_transcribe:

# Via API
curl -X POST https://<your-domain>/api/v1/serving/deploy_model \
  -H "Authorization: Bearer $TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"model_id": "<model-uuid>", "engine_name": "aws_transcribe"}'

Use the customer-facing HTTPS hostname for your deployment rather than localhost.

API Usage

Transcribe Audio (Batch)

curl -X POST "https://<your-domain>/runtime/models/<deployment-id>/v1/audio/transcriptions" \
  -H "Authorization: Bearer $TOKEN" \
  -F "file=@recording.wav" \
  -F "response_format=json"

Transcribe Audio (Streaming)

curl -X POST "https://<your-domain>/runtime/models/<deployment-id>/v1/audio/transcriptions" \
  -H "Authorization: Bearer $TOKEN" \
  -F "file=@recording.pcm" \
  -F "stream=true" \
  -F "language=en-US"

Streaming returns Server-Sent Events (SSE) with partial results.

Note: Streaming only supports pcm, ogg, and flac formats. The format is auto-detected from the file extension. If no language is specified, streaming defaults to en-US.

Request Parameters

Parameter	Required	Description
file	Yes	Audio file (multipart form-data)
response_format	No	text, json (default), or verbose_json
language	No	Override configured language
stream	No	true for streaming mode

Response Formats

json (default)

{
  "text": "Hello, how are you today?"
}

text

Hello, how are you today?

verbose_json

{
  "task": "transcribe",
  "language": "en",
  "duration": 3.45,
  "text": "Hello, how are you today?",
  "segments": [
    { "id": 0, "start": 0.0, "end": 1.2, "text": "Hello," },
    { "id": 1, "start": 1.3, "end": 3.4, "text": "how are you today?" }
  ]
}

Batch vs Streaming

Feature	Batch	Streaming
Audio formats	mp3, mp4, wav, flac, ogg, amr, webm	pcm, ogg, flac
Max duration	4 hours	5 minutes (300s)
Auto language detection	Yes	No (defaults to en-US)
Speaker labels	Yes	No
Custom vocabulary	Yes	Yes
PII redaction	Yes	No
Latency	Higher (job-based)	Real-time
Use case	Recorded audio	Live audio

Supported Languages

AWS Transcribe supports 100+ languages for batch and 30+ for streaming.

Auto-Detect (Batch Only)

Set language_code to auto (or leave empty) to enable automatic language identification. AWS Transcribe will detect the spoken language automatically.

{
  "language_code": "auto"
}

Note: Automatic language detection is only supported for batch mode. Streaming transcription requires an explicit language code and will default to en-US if not specified. To use a different language for streaming, pass the language parameter in your request.

Common Language Codes

Code	Language
auto	Auto-detect
en-US	English (US)
en-GB	English (UK)
es-ES	Spanish (Spain)
es-US	Spanish (US)
fr-FR	French
de-DE	German
ja-JP	Japanese
zh-CN	Chinese (Simplified)
pt-BR	Portuguese (Brazil)
ko-KR	Korean

Cost Considerations

Check AWS pricing for current rates.

Troubleshooting

"Missing external_endpoint configuration"

Ensure the model config includes external_endpoint with region and language_code.

"Unsupported media format"

Batch supports: mp3, mp4, wav, flac, ogg, amr, webm, pcm. Streaming supports: pcm, ogg, flac only.

"S3 access denied"

Verify IAM permissions include s3:PutObject, s3:GetObject, s3:DeleteObject for your bucket.

Streaming timeout

• Chunks must arrive within 30 seconds
• Total stream duration limited to 5 minutes
• Ensure audio data is sent promptly

Job timeout

Batch jobs timeout after 15 minutes by default. For longer audio, configure job_timeout_seconds in external_endpoint.