What you can do with KugelAudio that you could not do before, newest first. Each
entry links to the page that documents it.
kugel-3.5 is the default text-to-speech model when a request sends no
model_id. Send kugel-3 to stay on the previous model. See Models.- Creating a voice clone or adding reference audio to one now needs the
speaker’s consent: the required
speaker_consent field in the API and SDKs,
or a checkbox in the dashboard. See Speaker consent.
- Platforms that serve several end customers can create one project per
customer and issue and revoke its API keys over the API. See
Projects.
- Connect Claude, Cursor and other MCP clients to your KugelAudio account. See
MCP server.
- Organization owners and admins get one email when the organization’s credits
run out. See Errors.
- Speech enhancement with
clarity-1 is no longer a preview.
- One WebSocket connection can carry several audios, one after another. See
Several audios on one connection.
clarity-1 removes background noise from speech, or keeps one speaker’s
voice and removes every other voice and sound, for recordings and for live
audio. See Speech enhancement.- Every organization starts with a free month of Clarity, then switches to
paid. See Free month and paid plan.
- Python SDK 2.2.0 and JavaScript SDK 1.2.0 add speech enhancement. See the
Python and
JavaScript references.
- The LiveKit and Pipecat
integrations can remove background noise from incoming audio before your
agent hears it.
- Protect your account with an authenticator app or a passkey, sign in with a
passkey, and let your organization require two-factor authentication. See
Account security.
GET /v1/voices takes a language filter: de returns every German voice,
de-AT only Austrian German. See Voices.- The SDKs send
project_id, so a project’s pronunciation dictionaries apply
to SDK requests, including the Python LiveKit and Pipecat plugins. See
Dictionaries.
- The SDKs report their own failures to KugelAudio, and you can turn this off. See SDK diagnostics.
- The Python SDK requires Python 3.10 or newer.
- The Logs page of each project lists its text-to-speech requests with
their status, error code, cost and a trace ID to quote to support. See
Find a request in your logs.
- Register KugelAudio in Cognigy under the built-in Microsoft Azure Speech
Services vendor as an on-premises container. See
Cognigy.
- The LiveKit plugins take a
temperature option. See LiveKit.
- Use KugelAudio voices in Cognigy flows through Voice Gateway, with voice
style and text normalization settings. See Cognigy.
- The Python SDK’s
watermark extra checks a WAV file for the KugelAudio
watermark without a network call. See
Detecting the watermark.
- The LiveKit plugins take a
speed option (Python SDK 1.9.0, JavaScript SDK
0.10.0). See LiveKit.
- Codes with digits inside
<spell> tags are read in groups of four
characters with a short pause between groups, the way a person reads a
code aloud. See
Spell.
- All cloud text-to-speech audio carries an in-band watermark and a disclosure
header that mark it as AI-generated. See
AI-generated audio marking.
- Change
cfg_scale, speed or temperature between turns of a Python
streaming session with update_settings(). See
Python streaming.
- The JavaScript SDK includes a LiveKit Agents TTS plugin for TypeScript
agents. See LiveKit.
- Change voice settings between turns of a JavaScript streaming session with
updateSettings(). See JavaScript streaming.
- The Java SDK can choose pronunciation dictionaries per request.
- Speed up or slow down part of a text with
<prosody rate>. See
Per-span speed.
- Choose which pronunciation dictionaries a request uses. See
Choose dictionaries per request.
- Streaming sessions send a final frame after the last audio of a turn, so you
know the turn’s audio is complete (Python SDK 1.6.0, JavaScript SDK 0.8.0).
See Turn lifecycle.
- Request G.711 µ-law or A-law audio at 8 kHz for phone calls with
output_format. See Audio formats.
- WebSocket sessions report their audio time and cost when they close, so you
can bill your own customers per conversation. See
Per-session usage.
Last modified on October 16, 2026
