Speech Engines

VoiceBase offers three distinct proprietary speech engines, each tuned to address different use cases. The table below shows feature support by engine.

Titan

The Titan engine is VoiceBase’s default engine, offering the highest accuracy and best value for US English and Spanish content. Trained on call center audio, Titan is a great choice for US based customers working with agent/customer calls. It supports US English and Spanish only.

Europa

The Europa engine offers the highest accuracy for a wide range of regional accents for English and Spanish.

We recommend it for international use cases with combinations of English speakers from the UK, India, Singapore, and other regions.

No specific configuration is required when using its default language, US English. The default is the same for all regions and dialects of English.

Europa is trained mainly with European Spanish, as well as other variants. Europa Spanish may be configured as any one of the following: “language”:”es-ES”, “es-MX” or “es-US”. The resulting transcript will be the same regardless of which variant is configured.

Please note Europa’s price point is higher than our default engine, Titan.

Proteus

The Proteus engine is our general engine trained on a wide range of audio. All languages and features are supported with the Proteus engine.

Feature support by engine:

Feature Europa Proteus Titan
categories
custom vocabulary
number formatting
keywords-and-topics
keyword-spotting
languages en, es en-US, es-US
pci-ssn-pii-detection
pci-ssn-pii-redaction
stereo
swear word filter
verb-noun pairs
voice features  
voicemail    

Please note that number formatting may not be disabled on the Europa speech engine. Custom Vocabulary is supported with Europa, but the “Weighting” feature is not.

Configuration

Enable a specific speech engine by including the model parameter under speechModel when making a call to POST /media. Omitting this optional parameter defaults to the Titan engine for US-English, and Proteus for other languages.

{
  "speechModel": {
    "model": "Titan"
  }
}

Custom Speech Models

VoiceBase also offers the unique ability to train a custom speech model on a per customer basis in order to optimize transcription accuracy for a specific use case, lexicon, accent, or codec. Please reach out to your account manager to learn more about the custom speech model training process.

Language Options

Information about VoiceBase language offerings may be found here.