Audio craft: voice-first basics

Good audio writing starts with the ear: scripts are being shaped to sound natural when spoken, not just to read cleanly on a page. OmniVoice’s project page says punctuation and inline cues control pacing and expression, and its paper describes a text-to-speech system built for speech generation across more than 600 languages. (zhu-han.github.io, arxiv.org) That pushes basic craft choices to the front of the workflow. A spoken line needs shorter sentences, clearer syntax, and pauses that can be heard, because the listener cannot scan backward the way a reader can. (zhu-han.github.io, arxiv.org) The same logic applies to transitions. Instead of moving scene to scene with visuals alone, audio producers often use room tone, street noise, or a music cue to signal a shift before the next line arrives. (transom.org, insomnus.app) An ambient bed is the low, steady layer under a voice track — café clatter, traffic wash, air conditioner hum — and it works like background lighting in film. Mixing guides describe it as support, not competition: the bed should place the speaker somewhere without masking the words. (insomnus.app, fesliyanstudios.com) Music serves a different job. Under dialogue it usually stays restrained, and between scenes it can rise, breathe, and then clear space again so the next sentence lands cleanly. (fesliyanstudios.com, insomnus.app) That emphasis on clarity matches how larger platforms now talk about sound. Netflix’s engineering team said dialogue intelligibility can fail at several points in the production chain and built a measurement system to compare speech against music and effects in the final mix. (netflixtechblog.com) For small teams, that usually means choosing understandable speech over glossy processing. Heavy reverb, dense music, or aggressive effects can make a track feel polished while making key words harder to catch on laptop speakers or a phone. (netflixtechblog.com, insomnus.app) The production discipline behind that approach is deliberately lean. Audacity says its software is free, open source, and built for recording and editing audio, with support pages covering crossfades, loudness normalization, compression, limiting, noise reduction, and export. (audacityteam.org, support.audacityteam.org) As of April 20, 2026, Audacity’s download page lists version 3.7.7 for Windows, macOS, and Linux. Its support site also says the app is desktop-only, which keeps it in the role of a quick editor and dependable podcast workstation rather than a mobile capture tool. (audacityteam.org, audacityteam.org) The throughline is simple: write for a human voice, use sound to move the listener, and keep the mix clear enough that nobody has to rewind the sentence you needed them to hear. (zhu-han.github.io, netflixtechblog.com)