Gemini 3.8 Flash-Lite TTS

A product explanation, a lesson introduction, or a revised paragraph gives you something concrete to evaluate. Choose a passage that contains the vocabulary and sentence lengths you expect to use in the final recording. An extremely simple test may sound good without revealing how the voice handles your actual material. Keep the first sample short enough that you can listen to it several times and identify what needs changing.

Choose the voice before adding detailed direction. Neutral previews in the voice library help you compare vocal character, but they are not recordings of your script. Once you have a shortlist, generate the same passage with each candidate. Decide which voice makes the words easiest to follow, then adjust style or pace. Keep notes about the settings you prefer so later segments have a consistent starting point.

Gemini 3.8 Flash-Lite TTS uses the single-voice and two-speaker workflows available in this application. For one narrator, paste text into the main script field. For a dialogue, choose two voices and assign a speaker to each turn. A short voice description can explain the delivery you want, while scene direction gives the conversation shared context. Optional settings can remain empty when you want to hear the default interpretation.

Use supported tags sparingly. A short pause can separate an instruction from its explanation; a breath or sigh should belong to the performance rather than decorate every sentence. Too many cues make a script harder to review and can distract from the meaning. If the result does not match your intention, simplify the instructions before adding more. The prompting guide includes editable examples and explains where each kind of direction belongs.

The table below uses the application’s current pricing calculation. Both choices generate speech from text, expose the same thirty preset voices here, and offer WAV downloads after completion. Lite has the lower credit rate in this application. Flash is another option when you want to explore expressive narration and character exchanges. Neither a lower price nor a model name tells you which take will fit a particular scene.

For a fair listening comparison, keep the passage, voice, and instructions unchanged between requests. Listen to word stress, clarity, emotional consistency, and how the final sentence lands. Then change one variable at a time. Generation duration can vary with request conditions, so this page does not guarantee that one model will always finish first. A useful comparison concerns the audio you can actually use.

The estimate is based on started blocks of one hundred characters, rather than the number of English words or the final recording length. A script just over a block boundary can cost more than a slightly shorter one. Dialogue includes the submitted turns. The live estimate beside the generation button is the value to check after you edit text or switch models.

The example on this page is a short Mandarin reading generated with Lite for this site. Its language is labelled explicitly so you can distinguish it from the English Flash sample. It is useful as a real playback example, but it is not a matched comparison: the two clips use different text. Use your own repeated test passage if you want to compare models directly.

Once a task completes, listen through the result and download the WAV file. You can bring that file into an audio or video editor for trimming, volume adjustment, or placement alongside other clips. The studio does not automatically assemble a long project. Save meaningful filenames in your own project folder so you can recognise the model, voice, and version later.