Gemini TTS
A text-to-speech model in the Google Gemini series that supports multi-speaker voice configurations. It dynamically generates continuous audio featuring multiple roles and voice timbres within a single inference request, with voice timbres produced through model generation rather than audio concatenation.
Timeline (last 90 days)
Style instructions placed in the body text (such as reading-tone requirements) may be read aloud as body content by some TTS voices; the solution is to send only the text to be read to the model and pass style configuration via API parameters rather than the body.
The reader should receive only the message body and strip emoji first; otherwise emoji will be read as nonsensical syllables (e.g., the cake emoji 🎂 being read as "chha").
In the official demo, speaker Jack is mapped to the Orus voice and speaker Jill is mapped to the Zephyr voice.
Developers can complete multi-character voice generation within a single model call, without needing to make multiple separate requests and stitch the results together.
Firebase AI Logic announced support for Gemini's text-to-speech (TTS) model.
The voices provided by Gemini TTS are pre-trained acoustic models dynamically generated by the model at inference time, rather than audio splicing.
Developers prefix each line of text in the prompt with the corresponding speaker's name, and Gemini automatically switches voices to generate smooth multi-character dialogue.
Multi-speaker configuration is implemented by defining voice configurations that map each character name to a specific voice tone one-to-one.
All Facts (8)
The reader should receive only the message body and strip emoji first; otherwise emoji will be read as nonsensical syllables (e.g., the cake emoji 🎂 being read as "chha").
50%UnverifiedStyle instructions placed in the body text (such as reading-tone requirements) may be read aloud as body content by some TTS voices; the solution is to send only the text to be read to the model and pass style configuration via API parameters rather than the body.
50%UnverifiedIn the official demo, speaker Jack is mapped to the Orus voice and speaker Jill is mapped to the Zephyr voice.
50%UnverifiedDevelopers can complete multi-character voice generation within a single model call, without needing to make multiple separate requests and stitch the results together.
50%UnverifiedThe voices provided by Gemini TTS are pre-trained acoustic models dynamically generated by the model at inference time, rather than audio splicing.
50%UnverifiedDevelopers prefix each line of text in the prompt with the corresponding speaker's name, and Gemini automatically switches voices to generate smooth multi-character dialogue.
50%UnverifiedMulti-speaker configuration is implemented by defining voice configurations that map each character name to a specific voice tone one-to-one.
50%UnverifiedFirebase AI Logic announced support for Gemini's text-to-speech (TTS) model.
50%