Voice: TTS & STT

TomoriBot can send voice replies with text-to-speech (TTS) and understand audio messages through speech-to-text (STT). Choose speech providers in /providers or configure your own Google or OpenRouter speech models in /personal providers.

Text-to-speech

Ask a persona to send a voice message or use /generate voice-message. Enable voice messages under /config > Permissions and configure the persona’s voice under /config > Persona > Voice.

There are two TTS types:

• Voice design creates a voice from a written description, such as “a warm, soft-spoken voice.”
• Voice clone uses a reference recording to create a voice that resembles the sample.

Each type needs a compatible model. Save a description or assign a reference on the persona’s Voice page, where you can also choose which type to prefer. If the preferred option is unavailable, TomoriBot can use the other saved option. Descriptions are included in exported persona cards; voice reference assignments are not.

Tip

Use voice_design or voice_sample on /generate voice-message to try a voice for one message without changing the persona’s saved voice.

Providers and models

Google supports voice design, while OpenRouter offers voice design and cloning models. Configure models under /config > Models > Switch Models, or choose personal models under /personal config. Personal model failures may use the server’s model if server fallback is enabled. An explicitly selected one-off model does not fall back to another model.

ElevenLabs offers cloud voices and transcription. Add its key through /providers and assign a voice to each persona. Its free plan supports premade voices and has usage limits.

For local voice cloning and voice design, see the Text-to-Speech guides. Local cloning requires a reference sample, which you can upload under /config > Models > TTS Parameters & Voices.

Speech-to-text

A transcription model turns audio attachments into text TomoriBot can use in conversation. Choose it separately from TTS under /config > Models > Switch Models by picking a provider, then a model. Choose - No Model to stop transcription on the server’s key. Google, Vertex AI, and OpenRouter offer hosted transcription models, and adding one of their keys picks its default model when nothing else transcribes yet. OpenRouter lists many transcription models; register any of them by its model ID in /providers. Adding ElevenLabs also registers a transcription endpoint. For local options, see the Speech-to-Text guides.

You can also pick a personal transcription model under /personal config. Your own voice messages then use your key first and fall back to the server’s model if server fallback is enabled. Hosted transcription is billed by the provider for every voice message it transcribes.

You can hide visible transcripts under /config > Behavior > Notice Behavior > Speech Transcripts. The text still reaches the conversation context.