MumbleSynth: Procedural Gibberish Voices

Give every character a voice without recording a single line. MumbleSynth turns your dialogue text into gibberish speech, generated live in your game.

• Fully procedural: no audio clips, voices are generated from text at runtime
• 8 built-in voice presets, all customizable through Voice Profile assets
• 7 sound sources (glottal, reed, chip, whistle, growl, kazoo, chime) and 2 synthesis styles (formant speech and retro blips)
• 5 rhythm modes: per syllable, character, vowel, consonant or word
• 5 invented languages (Humanoid, Melodic, Guttural, Chirpy, Critter) plus a literal mode, with adjustable abstraction and vowel blur
• 7 emotions that change melody, rhythm, emphasis and tone
• Inline tags for emotion, pitch, speed and pauses; unknown tags such as TextMeshPro rich text are ignored
• Works with any language: English and other Latin scripts, Chinese, Japanese, Korean, Cyrillic and Greek are supported, and every other script falls back to a generic sound
• Every line plays at the same perceived loudness, whatever the voice or emotion
• Plays through a regular AudioSource, so mixer groups, volume and 3D spatial audio work as usual
• Live synthesis on the audio thread, plus a pre-rendered mode that's used automatically on WebGL
• Events for text reveal (counted the same way as TextMeshPro's maxVisibleCharacters), each spoken sound, mouth movement and line finished
• Native TextMeshPro integration
• Editor preview and WAV export from the Inspector
• Two demo scenes: a voice designer and a clickable low-poly village with five characters
• The voice engine itself has no render pipeline dependency. The village demo uses URP post-processing
• Full C# source included

MumbleSynth is a voice generator for the kind of gibberish speech you hear in cozy life sims, retro RPGs and cartoon platformers. You give it a line of dialogue, and it speaks it in a made-up language that follows the rhythm, punctuation and mood of your text. There are no audio files involved. Every voice is a handful of settings, so you can give fifty characters fifty different voices without bloating your build.

Questions go up at the end. Exclamations get more energy. Trailing "..." sentences slow down and fade out. Words in ALL CAPS get emphasis. The same word always sounds the same for a given character, so after a while their speech starts to feel like an actual language.

You can change how a line is delivered with seven emotions (happy, sad, angry, excited, scared, whisper and neutral), either for the whole line or mid-sentence with simple tags like <sad> or <pause=0.5>. Emotions change the melody, rhythm and tone of the voice, not just the pitch. Angry voices bark and punch the first word of every sentence. Sad ones get slow, low and breathy.

Eight presets get you started, from a chatty villager and a squeaky critter to a gruff grumbler, a wise elder, a retro text-box blip and a little robot. Every preset is fully editable, and you can audition changes right in the Inspector without entering Play mode.

Text reveal, mouth flaps and per-sound events are built in, so a typewriter dialogue box and a talking portrait take a few lines of code. Two demo scenes are included. One is a voice designer where you can tweak every parameter. The other is a small village with five characters you can click to chat with.