Identifying languages in multi-language audio Improving language identification accuracy Combining language identification with other Amazon Transcribe features Using language identification with batch transcriptions

Language identification with batch transcription jobs

Use batch language identification to automatically identify the language, or languages, in your media file.

If your media contains only one language, you can enable single-language identification, which identifies the dominant language spoken in your media file and creates your transcript using only this language.

If your media contains more than one language, you can enable multi-language identification, which identifies all languages spoken in your media file and creates your transcript using each identified language. Note that a multi-lingual transcript is produced. You can use other services, such as Amazon Translate, to translate your transcript.

Refer to the supported languages table for a complete list of supported languages and associated language codes.

For best results, ensure that your media file contains at least 30 seconds of speech.

For usage examples with the AWS Management Console, AWS CLI, and AWS Python SDK, see Using language identification with batch transcriptions.

Identifying languages in multi-language audio

Multi-language identification is intended for multi-lingual media files, and provides you with a transcript that reflects all supported languages spoken in your media. This means that if speakers change languages mid-conversation, or if each participant is speaking a different language, your transcription output detects and transcribes each language correctly. For example, if your media contains a bilingual speaker who is alternating between US English (en-US) and Hindi (hi-IN), multi-language identification can identify and transcribe spoken US English as en-US and spoken Hindi as hi-IN.

This differs from single-language identification, where only one dominant language is used to create a transcript. In this case, any spoken language that is not the dominant language is transcribed incorrectly.

Note

Redaction and custom language models are not currently supported with multi-language identification.

Note

The following languages are currently supported with multi-language identification: en-AB, en-AU, en-GB, en-IE, en-IN, en-NZ, en-US, en-WL, en-ZA, es-ES, es-US, fr-CA, fr-FR, zh-CN, zh-TW, pt-BR, pt-PT, de-CH, de-DE, af-ZA, ar-AE, da-DK, he-IL, hi-IN, id-ID, fa-IR, it-IT, ja-JP, ko-KR, ms-MY, nl-NL, ru-RU, ta-IN, te-IN, th-TH, tr-TR

Multi-language transcripts provide a summary of detected languages and the total time that each language is spoken in your media. Here's an example:


"results": {
        "transcripts": [
            {
                "transcript": "welcome to Amazon transcribe. ये तो उदाहरण हैं क्या कैसे कर सकते हैं ।一つのファイルに複数の言語を書き写す"
            }
        ],
    
    ...
    
        "language_codes": [
            {
                "language_code": "en-US",
                "duration_in_seconds": 2.45
            },
            {
                "language_code": "hi-IN",
                "duration_in_seconds": 5.325
            },
            {
                "language_code": "ja-JP",
                "duration_in_seconds": 4.15
            }
        ]
}

Improving language identification accuracy

With language identification, you have the option to include a list of languages you think may be present in your media. Including language options (LanguageOptions) restricts Amazon Transcribe to using only the languages that you specify when matching your audio to the correct language, which can speed up language identification and improve the accuracy associated with assigning the correct language dialect.

If you choose to include language codes, you must include at least two. There's no limit on the number of language codes you can include, but we recommend using between two and five for optimal efficiency and accuracy.

Note

If you include language codes with your request and none of the language codes you provide match the language, or languages, identified in your audio, Amazon Transcribe selects the closest language match from your specified language codes. It then produces a transcript in that language. For example, if your media is in US English (en-US) and you provide Amazon Transcribe with the language codes zh-CN, fr-FR, and de-DE, Amazon Transcribe is likely to match your media to German (de-DE) and produce a German-language transcription. Mismatching language codes and spoken languages can result in an inaccurate transcript, so we recommend caution when including language codes.

Combining language identification with other Amazon Transcribe features

You can use batch language identification in combination with any other Amazon Transcribe feature. If combining language identification with other features, you are limited to the languages supported with those features. For example, if using language identification with content redaction, you are limited to US English (en-US) or US Spanish (es-US), as this is only language available for redaction. Refer to Supported languages and language-specific features for more information.

Important

If you're using automatic language identification with content redaction enabled and your audio contains languages other than US English (en-US) or US Spanish (es-US), only the US English or US Spanish content is redacted in your transcript. Other languages cannot be redacted and there are no warnings or job failures.

Custom language models, custom vocabularies, and custom vocabulary filters

If you want to add one or more custom language models, custom vocabularies, or custom vocabulary filters to your language identification request, you must include the LanguageIdSettings parameter. You can then specify a language code with a corresponding custom language model, custom vocabulary, and custom vocabulary filter. Note that multi-language identification doesn't support custom language models.

It's recommended that you include LanguageOptions when using LanguageIdSettings to ensure that the correct language dialect is identified. For example, if you specify an en-US custom vocabulary, but Amazon Transcribe determines that the language spoken in your media is en-AU, your custom vocabulary is not applied to your transcription. If you include LanguageOptions and specify en-US as the only English language dialect, your custom vocabulary is applied to your transcription.

For examples of LanguageIdSettings in a request, refer to Option 2 in the AWS CLI and AWS SDKs dropdown panels in the Using language identification with batch transcriptions section.

Using language identification with batch transcriptions

You can use automatic language identification in a batch transcription job using the AWS Management Console, AWS CLI, or AWS SDKs; see the following for examples:

Sign in to the AWS Management Console.
In the navigation pane, choose Transcription jobs, then select Create job (top right). This opens the Specify job details page.
In the Job settings panel, find the Language settings section and select Automatic language identification or Automatic multiple languages identification.

You have the option to select multiple language options (from the Select languages dropdown box) if you know which languages are present in your audio file. Providing language options can improve accuracy, but is not required.
Fill in any other fields you want to include on the Specify job details page, then select Next. This takes you to the Configure job - optional page.
Select Create job to run your transcription job.

This example uses the start-transcription-job command and IdentifyLanguage parameter. For more information, see StartTranscriptionJob and LanguageIdSettings.

Option 1: Without the language-id-settings parameter. Use this option if you are not including a custom language model, custom vocabulary, or custom vocabulary filter in your request. language-options is optional, but recommended.


aws transcribe start-transcription-job \
--region us-west-2 \
--transcription-job-name my-first-transcription-job \
--media MediaFileUri=s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac \
--output-bucket-name DOC-EXAMPLE-BUCKET \
--output-key my-output-files/ \    
--identify-language \  (or --identify-multiple-languages) \
--language-options "en-US" "hi-IN"

Option 2: With the language-id-settings parameter. Use this option if you are including a custom language model, custom vocabulary, or custom vocabulary filter in your request.


aws transcribe start-transcription-job \
--region us-west-2 \
--transcription-job-name my-first-transcription-job \
--media MediaFileUri=s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac \
--output-bucket-name DOC-EXAMPLE-BUCKET \
--output-key my-output-files/ \
--identify-language \  (or --identify-multiple-languages)
--language-options "en-US" "hi-IN" \
--language-id-settings en-US=VocabularyName=my-en-US-vocabulary,en-US=VocabularyFilterName=my-en-US-vocabulary-filter,en-US=LanguageModelName=my-en-US-language-model,hi-IN=VocabularyName=my-hi-IN-vocabulary,hi-IN=VocabularyFilterName=my-hi-IN-vocabulary-filter

Here's another example using the start-transcription-job command, and a request body that identifies the language.


aws transcribe start-transcription-job \
--region us-west-2 \
--cli-input-json file://filepath/my-first-language-id-job.json

The file my-first-language-id-job.json contains the following request body.

Option 1: Without the LanguageIdSettings parameter. Use this option if you are not including a custom language model, custom vocabulary, or custom vocabulary filter in your request. LanguageOptions is optional, but recommended.


{
  "TranscriptionJobName": "my-first-transcription-job",  
  "Media": {
        "MediaFileUri": "s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac"
   },
  "OutputBucketName": "DOC-EXAMPLE-BUCKET",
  "OutputKey": "my-output-files/", 
  "IdentifyLanguage": true,  (or "IdentifyMultipleLanguages": true),
  "LanguageOptions": [
        "en-US", "hi-IN"  
  ]
}

Option 2: With the LanguageIdSettings parameter. Use this option if you are including a custom language model, custom vocabulary, or custom vocabulary filter in your request.


{
   "TranscriptionJobName": "my-first-transcription-job",  
   "Media": {
        "MediaFileUri": "s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac"
   },
   "OutputBucketName": "DOC-EXAMPLE-BUCKET",
   "OutputKey": "my-output-files/", 
   "IdentifyLanguage": true,  (or "IdentifyMultipleLanguages": true)
   "LanguageOptions": [
        "en-US", "hi-IN"  
   ],
   "LanguageIdSettings": { 
         "en-US" : { 
            "LanguageModelName": "my-en-US-language-model",
            "VocabularyFilterName": "my-en-US-vocabulary-filter",
            "VocabularyName": "my-en-US-vocabulary"
         },
         "hi-IN": {
             "VocabularyName": "my-hi-IN-vocabulary",
             "VocabularyFilterName": "my-hi-IN-vocabulary-filter"
         }
    }
}

This example uses the AWS SDK for Python (Boto3) to identify your file's language using the IdentifyLanguage argument for the start_transcription_job method. For more information, see StartTranscriptionJob and LanguageIdSettings.

For additional examples using the AWS SDKs, including feature-specific, scenario, and cross-service examples, refer to the Code examples for Amazon Transcribe using AWS SDKs chapter.


from __future__ import print_function
import time
import boto3
transcribe = boto3.client('transcribe', 'us-west-2')
job_name = "my-first-transcription-job"
job_uri = "s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac"
transcribe.start_transcription_job(
    TranscriptionJobName = job_name,
    Media = {
        'MediaFileUri': job_uri
    },
    OutputBucketName = 'DOC-EXAMPLE-BUCKET',
    OutputKey = 'my-output-files/',
    MediaFormat = 'flac',
    IdentifyLanguage = True,  (or IdentifyMultipleLanguages = True),
    LanguageOptions = [
        'en-US', 'hi-IN'
    ]
)

while True:
    status = transcribe.get_transcription_job(TranscriptionJobName = job_name)
    if status['TranscriptionJob']['TranscriptionJobStatus'] in ['COMPLETED', 'FAILED']:
        break
    print("Not ready yet...")
    time.sleep(5)
print(status)

Option 2: With the LanguageIdSettings parameter. Use this option if you are including a custom language model, custom vocabulary, or custom vocabulary filter in your request.


from __future__ import print_function
import time
import boto3
transcribe = boto3.client('transcribe')
job_name = "my-first-transcription-job"
job_uri = "s3://DOC-EXAMPLE-BUCKET/my-input-files/my-media-file.flac"
transcribe.start_transcription_job(
    TranscriptionJobName = job_name,
    Media = {
        'MediaFileUri': job_uri
    },
    OutputBucketName = 'DOC-EXAMPLE-BUCKET',
    OutputKey = 'my-output-files/', 
    MediaFormat='flac',
    IdentifyLanguage=True,  (or IdentifyMultipleLanguages=True)
    LanguageOptions = [
        'en-US', 'hi-IN'
    ],
    LanguageIdSettings={
        'en-US': {
            'VocabularyName': 'my-en-US-vocabulary',
            'VocabularyFilterName': 'my-en-US-vocabulary-filter',
            'LanguageModelName': 'my-en-US-language-model'
        },
        'hi-IN': {
            'VocabularyName': 'my-hi-IN-vocabulary',
            'VocabularyFilterName': 'my-hi-IN-vocabulary-filter'
        }   
    }
)

while True:
    status = transcribe.get_transcription_job(TranscriptionJobName = job_name)
    if status['TranscriptionJob']['TranscriptionJobStatus'] in ['COMPLETED', 'FAILED']:
        break
    print("Not ready yet...")
    time.sleep(5)
print(status)

Warning Javascript is disabled or is unavailable in your browser.

To use the Amazon Web Services Documentation, Javascript must be enabled. Please refer to your browser's Help pages for instructions.

Document Conventions

Identifying languages

Streaming language identification