I Gave My NPCs Voices Without Recording Anything

Summary

In this Mythfall development blog, solo developer UnitOfTime explains how he created procedurally generated mumbling voices for NPCs without recording audio. Mythfall is a browser-based MMO built from scratch in Go, and its dialogue changes frequently, making prerecorded voice lines impractical.

The system generates speech sounds as dialogue appears:

• Text timing provides the rhythm. Dialogue prints one letter at a time, with pauses adjusted for punctuation. Commas and sentence-ending marks give the speech a natural rhythm.
• Vowel sounds are synthesized. A script creates a buzzing source and shapes it by emphasizing three pitch bands for each vowel, drawing on older speech research. Breath-like noise, gradual fades, and small variations between buzzes make the result sound less artificial.
• Characters have distinct voices. Pitch and modeled vocal-tract length create differences in pitch and timbre. Older characters get a slow pitch oscillation; making the pitch more random sounded drunk rather than old.
• Text determines when and what to play. Rather than producing a sound for every letter, the game plays a vowel sound when a vowel group begins. The letter selects the vowel, keeping the number of sounds manageable and repeated lines consistent. Slightly overlapping sounds create a continuous stream of chatter.
• Punctuation adds expression. Pitch follows phrases and resets at commas. Questions rise at the end, exclamations begin high, ellipses make the voice quieter, and capitalized words are treated as shouting. These effects are generated from the dialogue itself rather than authored for each line.
• Small extras use the same text timer. A line can trigger a sound effect at a particular letter, such as a cat’s purr. If the player reveals the whole line at once, the voice stops because the letters stop printing.

The developer also experimented with noisy consonant clicks but removed them because they sounded like clicking. The vowel sounds worked better on their own. Completed character voices use about half a megabyte, and adding a voice to a character takes one line of code.

Key Takeaways

The developer’s main conclusions are that rhythm matters more than individual sounds, small imperfections make synthetic audio feel more human, and punctuation can add expressive melody without requiring dialogue-specific work.

Guide and Tutorial Content

The video is a technical walkthrough of the voice system, covering text-box timing, basic voice synthesis, character differentiation, text-driven sound selection, and punctuation-based expression.

Main Speaker and Source

• UnitOfTime — Mythfall’s developer and the video’s narrator.