Text read aloud by a voice you pick. You type the text in the panel, so you can hear the price before you commit.
Entry I
Run /speech, optionally with a model (default Kokoro 82M, from 1 credit). The panel opens empty: press Add text, paste up to 2000 characters, then pick the model family, model and Voice. MAI-Voice-2, MAI-Voice-2 Flash, MAI-Voice-2.1 and MAI-Voice-2.1 Flash also offer a Style: Cheerful, Sad, Excited, Angry or Neutral.
The price follows the length of your text and is shown on Generate. Credits are taken only when you press Generate. The button shows the price for the settings you picked, and your free 25 credits a day are spent before paid ones. A run that fails is refunded.
Entry II
Commas and full stops set the rhythm. An ellipsis (…) gives a pause or hesitation; a question mark lifts the end.
"Forty-two dollars fifty" instead of "$42.50"; read phone numbers digit by digit; write "Doctor" for "Dr.". Smaller, faster models expand these less reliably.
Write the feeling into the sentence, or pick a Style on the models that have one. Some models read stage directions aloud, so keep them out unless the model supports tags.
Your text is sent as you write it. Eleven v4, Eleven v4 Turbo, Eleven v3 and Eleven v3 Conversational understand short audio tags such as [whispers], [laughs] or [sighs]. Test them with your voice; some suit some voices better.
Entry III
Enhance has a small chat model (Claude Haiku 5.5, GPT-6 Luna, MiMo v2.6 Flash or DeepSeek V4.1 Flash 1017) rewrite your prompt for the model you picked, for +2 credits. You see both versions and choose Using: Yours or Using: Enhanced, and you can edit the enhanced one before generating. For speech it fixes grammar and punctuation for a natural read, keeps your meaning, stays near the original length and adds no tags you did not use.
Entry IV
What you can type after /speech, read from the command itself. You can run it again after 10 seconds.
| Option | What it does | Takes |
|---|---|---|
model |
TTS model (type to search) | texttype to search |
hidden |
Only you can see this (default: false — post in channel) | true / false |
Entry V
Requests you can copy into Discord and change to fit.
/speech → Add text: Good evening, everyone! Game night starts in ten minutes… grab a snack, and meet us in the Lounge channel.
Why it works. Short sentences, an ellipsis for a pause, and the time written in words.
/speech model: (Eleven v3 or v4) → Add text: [whispers] Do you hear that? … Something is moving downstairs.
Why it works. An audio tag on a model that reads tags, plus a pause.
/speech model: (a MAI-Voice model) → Add text: Congratulations, you did it! → Style: Cheerful
Why it works. The Style menu sets the delivery on the models that have it.
Entry VI
Paste the final text; every character is priced.
Try two voices on one sentence before reading a long text.
For long scripts, break at full stops. Very long sentences sound breathless.
Entry VII
Up to 2000 characters per run.
Up to 2 speech jobs at once, 10 seconds between commands, 2 minutes per run.
Each model has its own voices; switching model resets the voice to that model's default.
Entry VIII
Speech is priced from the length of the text, with a minimum per model.
Not in /speech. To make a photo talk in your own voice, attach your recording to /clip avatar.
No. For singing use /music.
Prompting advice is from these free public guides, matched to what ALMA lets you set. Prices and limits are read from the bot itself, as of 22 October 2026.
Type /speech in any server with ALMA, or in a direct message to the bot.
Something here wrong, or missing? Tell us.
