Skip to navigation
FeaturesTTS Vendor Settings

KugelAudio

Speed, guidance, temperature, normalization and pronunciation dictionaries for KugelAudio TTS.

KugelAudio tuning is passed in synthesizer.options.

{
  "verb": "say",
  "text": "Thanks for calling, how can I help you today?",
  "synthesizer": {
    "vendor": "kugelaudio",
    "language": "en-US",
    "voice": "1071",
    "options": {
      "speed": 1.0,
      "cfg_scale": 2.0,
      "normalize": true
    }
  }
}

You can also set these in the portal under Extra Options on the speech credential. Credential settings apply to streamed TTS, such as an agent’s replies, with settings on the verb taking precedence; a say verb uses only the options on the verb.

The voice value is a KugelAudio voice id, which is numeric (for example 1071). The model defaults to kugel-3.

Options

speed
number

Speaking rate. Defaults to 1.0. Values above 1.0 are faster, below are slower.

cfg_scale
number

Guidance scale — how closely the read follows the voice. Defaults to 2.0.

temperature
number

Sampling temperature. Lower values give a more consistent read; higher values more variation between renders of the same text.

normalize
boolean

Expand numbers, dates and abbreviations before synthesis. Defaults to true. Normalization follows the request language.

project_id
number

KugelAudio project the request is billed and filed under.

dictionary_ids
array

Ids of the pronunciation dictionaries to apply, as an array of numbers. An empty array [] means “no dictionaries”, which is different from leaving the field out.

Language

jambonz sends only the primary subtag of language (de-DE is sent as de), and only when KugelAudio supports it. An unsupported or empty language is left out, so the voice’s own language applies.

Endpoint override

Set api_uri in options (on the credential’s Extra Options it applies to streamed TTS only) to use a different KugelAudio host, for example api.eu.kugelaudio.com to pin the EU region. Leave it out to use api.kugelaudio.com.

Streaming

KugelAudio supports TTS streaming. It returns word timestamps, so when a caller barges in on an agent, the assistant turn is trimmed to what the caller actually heard.