Narrate your newsletter with Gemini 3.8 Flash TTS

AI Studio's speech page lets you hear every change before you write any code.

1. 1Go to aistudio.google.com/generate-speech and sign in with your Google account.
2. 2In the model menu, pick Gemini 3.8 Flash TTS.

You should see the speech page with Gemini 3.8 Flash TTS picked and an empty box for your script.

Flash TTS gives the richest voice for long reads. Flash-Lite TTS suits high volume, and the two take the same scripts.

A designed voice sounds the same in every issue, so you describe your narrator only once.

1. 1Open the Voice Design studio from the voice options.
2. 2Describe your narrator in 1 or 2 sentences: age, timbre, accent and how they speak. Use the example below as a start.
3. 3Play the preview, then save the voice with a name such as Newsletter narrator.

A clear, warm narrator in her 30s with a neutral American accent, reading the news at a steady, friendly pace.

You should see your new voice saved in your voice list, with a preview you can play.

Prefer a ready-made voice? Of the 30 built-in voices, Google labels Charon "Informative" and Sulafat "Warm".

Gemini reads your text word for word, so anything left in the script gets spoken.

1. 1Paste one section of your issue into a plain text editor.
2. 2Delete links, image captions, button text and emoji.
3. 3Add <short pause> between stories and <long pause> before a new section.
4. 4Write one key word in capitals where you want the narrator to stress it.

Good morning. Here are 3 stories worth your time. <short pause>

First, [STORY 1 IN 2 OR 3 SENTENCES]. <short pause>

Second, [STORY 2 IN 2 OR 3 SENTENCES]. This one MATTERS if you [WHO IT AFFECTS]. <long pause>

That's it for today. See you tomorrow.

You should see a clean script with only spoken words and a few tags in angle brackets.

Make one script per section. A single request takes up to 8,192 input tokens, and short parts are easier to redo.

One short style note sets the mood for a whole section, and Google says most scripts need none.

1. 1Pick your narrator voice and paste your script.
2. 2Click Run with no style note first, and listen.
3. 3If the read needs a mood, add a style note of a few words, such as warm and upbeat, and run it again.
4. 4When it sounds right, download the audio.

You should see your section read aloud in your narrator's voice, with pauses where you put the tags.

Keep fixed traits such as age or accent out of the style note. Those belong in the voice design.

With your voice ID and one command, each new issue becomes audio without opening AI Studio.

1. 1Copy your narrator's voice_ ID from AI Studio. Click Get code to see the request for your own setup.
2. 2On the API keys page, click Create API key. In your terminal, run export GEMINI_API_KEY="[YOUR KEY]".
3. 3Save your script as script.txt in the folder you run the command from.
4. 4Put your voice ID into the command below and run it.

jq -n --rawfile text script.txt --arg voice "[YOUR VOICE ID]" \
  '{model: "gemini-3.8-flash-tts", input: [{type: "user_input", content: [{type: "text", text: $text, annotations: [{type: "speech_metadata", style: "warm and clear"}]}]}], response_format: {type: "audio"}, generation_config: {speech_config: [{voice: $voice}]}}' \
  | curl -s -X POST "https://generativelanguage.googleapis.com/v1beta/interactions" \
    -H "x-goog-api-key: $GEMINI_API_KEY" -H "Content-Type: application/json" -d @- \
  | jq -r '[.steps[] | select(.type=="model_output") | .content[] | select(.type=="audio")] | last | .data' \
  | base64 --decode > issue.wav

You should see an issue.wav file that plays your section in your narrator's voice.

Keep your API key in the environment variable. Never paste it into a script you share or commit. jq handles quotes and line breaks in your script for you.