Skip to main content
The voice pipeline is now multi-provider. You can choose from five text-to-speech engines, six speech-to-text models, and two noise suppression engines, all configured in a redesigned Voice Studio experience.

​
Text-to-Speech Providers

Assistable supports five TTS providers, each with its own model lineup and tuning controls. Rime is the recommended provider for most use cases. Models:
  • Coda (new default): flagship conversational model.
  • Natural: balanced quality and latency.
  • Natural HD: higher fidelity.
  • Arcana v3: expressive character voices.
  • Kokoro: multilingual support.
All Rime models support Expressive Mode (context-aware emotion) and Language Boost across 11 languages.

​
ElevenLabs

ElevenLabs offers the full model family with fine-grained tuning:
  • Eleven v2: flagship quality.
  • Flash v2 / v2.5: low latency.
  • Turbo v2 / v2.5: ultra-low latency.
  • Multilingual v2: 29 languages.
Tuning controls: Stability, Similarity Boost, Style, Speaker Boost.

​
MiniMax

MiniMax models:
  • speech-2.8-turbo: fast conversational speech.
  • speech-2.5-hd: high-definition audio.

​
Amazon Polly

AWS Polly provides a wide range of voices across 30+ languages with Neural and Standard engines.

​
Microsoft Azure

Azure TTS offers locale-precise voices with automatic lowest-latency region selection.

​
Inworld

Inworld models:
  • Mini: lightweight conversational model.
  • Max: higher expressiveness.

​
Resemble

Resemble models:
  • Turbo: fast synthesis.
  • Rapid: ultra-low latency.

​
Speech-to-Text Models

Your assistant’s hearing is now configurable. Choose from six transcription engines. Deepgram Flux is the new recommended default. It is a conversation-native model with fine-grained end-of-turn controls, so the assistant knows exactly when the caller has finished speaking. Supports 10 languages plus auto-detect and multilingual code-switching.

​
Deepgram Nova-3

Nova-3 supports 80+ languages and locales with Smart Format (automatic punctuation, capitalization, numeral conversion) and Keyterm Boost (bias transcription toward your product names and industry jargon).

​
Deepgram Nova-2

Nova-2 offers broad language coverage with Smart Format and numeral conversion.

​
AssemblyAI Universal-Streaming

Streaming transcription with configurable turn-silence thresholds. Supports 6 languages plus auto-detect.

​
Azure Fast

Locale-precise transcription with automatic lowest-latency region selection.

​
Noise Suppression

Noise suppression cleans up caller audio before it reaches transcription. Enable it for calls from cars, cafés, and job sites. Industry-leading voice isolation. Krisp removes background noise while preserving caller speech.

​
DeepFilterNet

DeepFilterNet offers adjustable attenuation limits and an advanced mode for full control over how aggressively noise is removed.

​
Voice Studio UI

The Voice Studio experience is redesigned around three workspaces:

​
Voice Settings Panel

The Voice Settings panel is your central configuration workspace. It includes:
  • Provider and model selection (TTS and STT).
  • Voice tuning controls (Stability, Similarity Boost, Style, Speaker Boost for ElevenLabs; Expressive Mode and Language Boost for Rime).
  • Noise suppression engine selection and configuration.
  • Interruption and turn-taking controls (end-of-turn thresholds, silence detection).
  • Auto-save: a live Saving indicator appears as you edit, then changes to Saved. Nothing goes live to callers until you press Publish.

​
Voice Library

Browse the multi-provider voice catalog:
  • Instant search: search by voice name, voice ID, or accent.
  • Filters: filter by provider, language, or accent.
  • Curated tabs: Recommended (staff-curated voices) and Favorites (your starred voices).
  • One-click previews: play any voice inline.
  • Community voting: upvote voices to help surface the best options.

​
Compare Dock

Shortlist up to 8 voices and compare them side by side:
  • Back-to-back playback: play voices in sequence with auto-advance.
  • Adjustable speed: 0.75× to 2× playback speed.
  • One-click apply: select the winner and apply it to your assistant.

​
My Voices

Manage custom voices:
  • Import a recording: upload an audio file to clone a voice.
  • Clone from a sample: provide a short sample and generate a custom voice.
  • Duplicate by ID: copy a voice by its ID.
Custom voices are private to your workspace.

​
How to Configure Voice Settings

1

Open Voice Settings

Open any assistant in your portal and navigate to the Voice Settings panel.
2

Choose a TTS provider and model

Click Change voice to open the Voice Library. Browse, search, or filter to find a voice. Play previews, add voices to the Compare dock, and select the one you want.
3

Choose a transcription model

In the Voice Settings panel, select a speech-to-text model from the dropdown. Configure end-of-turn thresholds, Smart Format, or Keyterm Boost if the model supports them.
4

Enable noise suppression

Toggle noise suppression on and choose Krisp or DeepFilterNet. If using DeepFilterNet, adjust the attenuation limit and enable advanced mode if needed.
5

Tune voice parameters

If your provider supports tuning (ElevenLabs, Rime), adjust Stability, Similarity Boost, Style, Speaker Boost, Expressive Mode, or Language Boost.
6

Publish

Press Publish to make the changes live on calls. The assistant will use the new voice, transcription model, and noise suppression settings on the next call.

​
Notes

Voice changes do not affect calls in progress. The new settings apply to calls that start after you press Publish.
Custom voices (cloned or imported) are private to your workspace and do not appear in the public Voice Library.
Noise suppression adds a small amount of latency (typically under 100ms). If you are optimizing for the lowest possible latency, test with and without noise suppression to measure the impact.