Sound Effects & Speech
Generate game sound effects, ambient loops and expressive character dialogue through the 3D AI Studio MCP connector, with prompting tips, voices, audio tags, WAV output and costs.
Your assistant can create sound effects and spoken dialogue for your projects: footsteps, impacts, UI clicks, ambient loops, NPC lines, narration and more. This makes it easy to build a game with an AI agent: the agent writes the code, generates the 3D models, and now also creates the sounds and voices and drops them into your project.
Both tools spend your normal credits. Every result is saved to your dashboard and cleared for commercial use.
At a glance
| Tool | What it makes | Cost |
|---|---|---|
Sound effectsgenerate_sound_effect | Sound effects, ambiences, seamless loops and short musical stingers from a description, 0.5 to 22 seconds | 1 credit per started 2 seconds (5 seconds = 3 credits) |
Speechgenerate_speech | Expressive speech in 90+ languages with 21 voices, directed with audio tags | 1 credit per started 100 characters |
You can ask for 1 to 4 variations of the same request. Each variation costs the same as a single result, so 3 versions of a 5 second sound cost 9 credits.
Audio needs a new permission. If you connected your assistant before audio was available, it will ask you to reconnect the first time it tries to make a sound. Connect again from your assistant and approve Generate sound effects and speech using your credits. Your other tools keep working either way. See Troubleshooting.
Where the result shows up
- Claude, ChatGPT and other chat apps with interactive cards show a player right in the conversation, with one player per variation.
- Coding agents (Claude Code, Cursor, Codex, VS Code, OpenCode) receive a direct file link and can download the file straight into your project, for example into
assets/audio/. - Every client can open the result in your dashboard.
MP3 or WAV
| Format | Use it for |
|---|---|
| MP3 (default) | Previews, websites, quick listening. Small files |
| WAV | Game engines and anything that loops. Unity, Unreal and Godot import WAV directly, and WAV loops play without the small gap MP3 can leave at the loop point. About 10 times larger |
Both formats cost the same. Ask for it in plain words: "as WAV for my game".
Sound effects
How to describe a sound
Describe the sound itself, not the scene or the story. A good description names what makes the sound, what happens, the material or surface, and the space it happens in.
| Instead of | Try |
|---|---|
| "a scary castle" | "heavy oak door creaking open slowly, stone hallway reverb" |
| "a battle" | "two steel swords clashing, sharp metallic ring" |
| "a menu sound" | "soft UI click, short and clean" |
| "rain" | "steady light rain on leaves, seamless loop" |
Tips that make a big difference:
- Use sound design words: impact, whoosh, riser, braam (a big cinematic hit), glitch, drone, ambience, foley, stinger, one shot.
- Describe sequences in order: "footsteps on gravel, then a metal gate slams shut". For exact timing, make the parts separately.
- Music works for short pieces: name the style, instrument and tempo, for example "90s hip hop drum loop, 90 BPM" or "vintage brass stab in F minor". It is not meant for full songs, vocals or dialogue.
- Describe characteristics, not famous sounds. "Humming energy blade with a low buzz" works; naming a film or franchise does not.
- One sound per request. For sounds a game repeats (footsteps, hits, pickups), ask for 3 or 4 variations so the game does not repeat the same clip.
Length and loops
| Sound | Good length |
|---|---|
| UI clicks, impacts, hits | 0.5 to 2 seconds |
| Whooshes, stingers, pickups | 1 to 4 seconds |
| Ambiences and background loops | 10 to 22 seconds |
A length that is too long for a short sound adds silence or repetition. Ask for a seamless loop for ambiences, engine hums and background beds, and get it as WAV.
Example prompts
- "Make 4 variations of footsteps on wet stone for my player character, as WAV."
- "Create a 20 second seamless forest ambience loop with birds and wind."
- "Generate a short coin pickup sound and a level complete stinger."
- "Make a heavy sci-fi door opening sound with a hydraulic hiss."
Speech and dialogue
One voice per request
Each speech request is spoken by one voice. For a dialogue between characters, your assistant makes one request per line (or per group of lines from the same character) and keeps the same voice for each character throughout. Speaker names like "GUARD:" are never written into the text, because the voice would read them out. Each line gets its own name, such as guard_03_warning, so the files stay in order.
Directing the performance with audio tags
Write short directions in square brackets right before the words they affect. They shape the delivery and are not spoken.
| Kind | Examples |
|---|---|
| Emotion | [excited] [sarcastic] [curious] [nervously] [angry] [sad] [mischievously] [deadpan] |
| Voice quality | [whispers] [shouting] [low, gravelly voice] [quietly, with controlled fear] |
| Sounds a person makes | [laughs] [chuckles] [sighs] [exhales sharply] [clears throat] [gulps] [crying] |
| Pacing | [short pause] [long pause], an ellipsis (...) for hesitation, a dash for an interruption |
| Accent (experimental) | [strong French accent] |
Example dialogue (two requests):
Voice Callum: "[low, gravelly voice] Keep the lantern covered. [short pause] If they see the light... they'll know we crossed the river."
Voice Lily: "[quietly, with controlled fear] I heard them at the bridge. Not soldiers. [gulps] Something else."
Tips:
- Be explicit. A vague tag can be performed as a sound effect instead of a way of speaking.
[low, gravelly voice]works better than a single word. - Match tags to the voice. A calm narrator voice will not shout convincingly.
- Use one to three tags per sentence. Capital letters add emphasis: "It was a VERY long day."
- Names and invented words: your assistant can write the pronunciation in phonetic symbols between slashes, for example "/ˈkeɪlən/", for a name like Kaelen.
- Numbers, prices and dates are read more reliably when written out in words.
- Long scripts are split into requests of up to 5,000 characters, with the same voice and settings for each part.
Voices
| Voice | Character |
|---|---|
| Rachel (default) | Female, American, calm and clear. Narration |
| Aria | Female, American, expressive. Characters |
| Sarah | Female, American, soft and reassuring |
| Laura | Female, American, upbeat and quirky |
| Jessica | Female, American, expressive and young |
| Matilda | Female, American, warm and friendly |
| Alice | Female, British, confident |
| Lily | Female, British, warm. Narration, audiobooks |
| Charlotte | Female, Swedish accent, smooth. Characters |
| Roger | Male, American, confident |
| Liam | Male, American, articulate and young |
| Will | Male, American, friendly and young |
| Eric | Male, American, smooth and friendly |
| Chris | Male, American, casual |
| Brian | Male, American, deep and resonant. Trailers |
| Bill | Male, American, older and trustworthy |
| George | Male, British, warm and mature. Storytelling |
| Daniel | Male, British, authoritative. Announcements |
| Charlie | Male, Australian, natural and casual |
| Callum | Male, transatlantic, gravelly and intense. Game characters |
| River | Non binary, American, calm and neutral |
You can also ask for a stability setting: lower values give a more expressive, tag driven performance, higher values a steadier narration. Ask for 2 or 3 variations of important lines to pick the best take.
Example prompts
- "Write and voice 5 short barks for a grumpy blacksmith NPC, one file each, as WAV."
- "Voice this tutorial script with a friendly female voice."
- "Create a dialogue between a nervous scout and a stern captain for my intro scene."
- "Narrate this paragraph in German."
Example workflow: a game with sound
"I'm building a small dungeon crawler in Godot. Generate 3 variations of footsteps on stone, a sword swing, a torch crackle loop and a door creak as WAV, plus voice lines for the gatekeeper NPC, and put them into res://assets/audio."
A coding agent generates each sound, downloads the files into your project and can wire them into your scenes.
Commercial use
Sounds and speech you generate can be used commercially, including in games and videos you sell. A few rules come from the model provider and always apply:
- Do not imitate the voice of a real person, or make it sound like a real person said something they did not.
- Say that a voice is AI generated wherever the law requires it.
- Do not use the results to train other AI models.
Frequently Asked Questions
Not through the connector. It uses the 21 built in voices listed above.
Each speech request uses one voice. One file per line also makes it easy to trigger each line separately in a game.
Ask for the sound again as WAV. MP3 files can leave a tiny gap at the loop point in some game engines.
In your dashboard, next to your models and images. Your assistant can also find them: "show me the sounds I made today".
Images & Uploads
How to give your AI assistant a photo, sketch or reference image for 3D generation or editing: the inline upload card, the upload page, links and library images.
Credits & Limits
How the 3D AI Studio MCP connector spends credits: website prices, personal or team credits, confirmations for large requests, refunds, rate limits and plan needs.