Supported languages

Language support for Medallia Speech

Speech supports transcription of voice data in several languages. Speech uses two separate transcription engines to maximize language support: Medallia's in-house transcription engine (Engine 1) and Amazon Transcribe (Engine 2).

Supported language matrix

Medallia Speech supports a variety of languages and regional dialects, with different transcription engines and features available for different languages.

Note:

  • Languages listed as Engine 1 use Medallia's in-house transcription engine, and those listed as Engine 2 use the Amazon Transcribe service.

  • We separate Engine 1 languages into quality tiers. For a description of each language quality tier in the following table, see the Language tiers section below.

Consult the table below to view languages and their supported features in Medallia Speech.

LanguageEngineFeature supportText Analytics

Arabic (Gulf) / ar-AE

2
  • *Custom vocabulary
  • Substitution
done

Chinese (Simplified) / zh-CN

 2
  • *Custom vocabulary
  • Substitution
done

Chinese (Traditional) / zh-TW

 2
  • *Custom vocabulary
  • Substitution
done

Czech / cs-CZ

 2
  • *Custom vocabulary
  • Substitution
done

Danish / da-DK

 2
  • *Custom vocabulary
  • Substitution
done

Dutch (Europe) / nl-NL

 1 (Tier 3) and 2 For Engine 1 only:

  • Acoustic emotion

  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

English (Australia) / en-AU

1 (Tier 1)

  • Acoustic emotion

  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

English (Great Britain) / en-GB

1 (Tier 1)

  • Acoustic emotion

  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

English (United States) / en-US

1 (Tier 1)

  • Acoustic emotion

  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

French (Canada) / fr-CA

1 (Tier 2)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

French (France) / fr-FR

1 (Tier 2)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

German (Germany) / de-DE

1 (Tier 2)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

Greek / el-GR

 2
  • *Custom vocabulary
  • Substitution

Hebrew / he-IL

 2
  • *Custom vocabulary
  • Substitution
done

Hungarian / hu-HU

 2
  • *Custom vocabulary
  • Substitution
done

Italian (Italy) / it-IT

1 (Tier 2)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

Japanese / ja-JP

 2
  • *Custom vocabulary
  • Substitution
done

Korean / ko-KR

 2
  • *Custom vocabulary
  • Substitution
done

Norwegian / no-NO

 2
  • *Custom vocabulary
  • Substitution
done

Polish / pl-PL

 2
  • *Custom vocabulary
  • Substitution
done

Portuguese (Brazil) / pt-BR

1 (Tier 1)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

Portuguese / pt-PT

 2
  • *Custom vocabulary
  • Substitution
done

Romanian / ro-RO

 2
  • *Custom vocabulary
  • Substitution
done

Slovak / sk-SK

 2
  • *Custom vocabulary
  • Substitution
done

Spanish (Spain) / spa-ES

1 (Tier 1)
  • Acoustic emotion
  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

Spanish (Americas) / spa-AMER

1 (Tier 1)
  • Acoustic emotion

  • Custom vocabulary

  • Music detection

  • Numeric redaction

  • Substitution

done

Swedish / sv-SE

 2
  • *Custom vocabulary
  • Substitution
done

Turkish / tr-TR

 2
  • *Custom vocabulary
  • Substitution
done

* Custom vocabulary (OOV) for Engine 2 has limited availability. It is configured differently than Engine 1 and must be enabled by a Medallia expert before use.

Available features

The following list defines the features available for the designated languages in the table below:

  • Acoustic emotion — Discover changes in customer's emotional expressions throughout an interaction.

  • Custom vocabulary (OOV) — Transcription accuracy can be improved by recognizing brand- and industry-specific terms.

  • Music detection — Detect when utterances should be classified as background music and not transcribed.

  • Numeric redaction — Numeric phrases are redacted from audio and transcripts.

  • Substitution — Correct errors in transcripts automatically using rules that find and replace errors with corrected text.

  • Voice anonymization — Audio can be modified to mask speakers' identities.

Note: Translation services are available if needed. To learn more, see Translations and Translating comments.

Language tiers

We separate Engine 1 languages into tiers to differentiate the datasets used for training language models. Higher tiers feature more robust models that consistently produce higher quality results.

Tier 1

Tier 1 languages are available to use with Speech. Tier 1 language models have been trained with a robust dataset and produce high quality transcription results. Audio and transcription analysis is recommended for output tuning prior to using Tier 1 languages in production environments.

Tier 2

Tier 2 languages are available to use with Speech. Tier 2 language models have been trained with a limited dataset, meaning transcription quality may not meet production requirements. Audio and transcription analysis is strongly recommended to evaluate transcription output prior to using Tier 2 languages in production environments.

Tier 3

Tier 3 language models are not generally available and are considered beta models. Tier 3 languages are not available for Speech without approval from Medallia Product and Engineering teams, and this approval process must include thorough audio and transcription analysis.

Important: Speech implementations using Tier 3 languages may incur significant additional lead time.

Text Analytics support

Ensure Text Analytics is turned on for the languages used for voice data by your company. On the Reporting > Text Processing > Global Text Analytics Settings screen in Medallia Setup, review the language processing settings for the languages used by your company, and make any changes as needed. For more information, see Text Analytics supported languages.

Note: For each language you use for Speech, configure whether to process comments in the native language or its translation. If you choose to process the native language, Text Analytics processes transcribed call records only when the native language is English, French, or Spanish. For example, if you configure Text Analytics to process Italian in the native language, Text Analytics does not process transcribed Italian call records. For more information, see Global Text Analytics Settings.

Amazon Transcribe languages

Speech uses Amazon Transcribe to support the languages listed as Engine 2 in the table above.

Some Speech features, including acoustic emotion, custom vocabulary, and redaction are not available when using Amazon Transcribe languages. Substitution is not enabled for Amazon Transcribe languages (Engine 2) by default. If you wish to enable substitution for these languages, contact a Medallia expert.

Amazon Transcribe's Language Identification feature can be used to identify languages supported with Amazon Transcribe and Medallia Text Analytics. One language is transcribed for any given call. Language identification is not available for Medallia's in-house transcription engine. For more information, see Connector settings.

Amazon Transcribe language models are developed and trained by Amazon. For more information, see Amazon Transcribe.

Important: Medallia Speech has altered our acoustic emotion calculation to comply with new regulations in the EU per the EU AI Act. Currently, acoustic emotion is calculated using text-based emotion recognition only, and voice-based processing to determine emotion is discontinued.