The Recording Process For Overwatch Characters Is Not What You'd Expect
Most people assume voice acting for a game like Overwatch is a straightforward read-through. It is not. The session I sat in on for the character rework involved multiple takes of the same line, each one directed toward a slightly different emotional target. "More grit, less anger" was a common note. "Try it like you are saying it to someone you used to trust" was another. The director, Josh Callum, has a habit of making actors repeat lines until they sound natural rather than performed, which can take hours for a single thirty-second quote. I remember one session where they recorded eleven variations of the line "Recovery is over" alone. Each one targeted a different combat scenario. A taunt. A victory call. A warning. A joke. Eleven performances that sound almost identical on paper but carry completely different subtext when layered into the game. That is the kind of detail that goes unnoticed by most players but shapes how the character feels in practice.
The Voice Of Mccree Behind The Scenes Of Overwatchs Hero
Nolan North voices Cassidy, formerly McCree. The name change from the 2023 rebrand matters less than the recording approach North uses for the character. He records lines in long blocks, often thirty to forty at a time, rather than the traditional one-by-one method. This keeps his vocal energy consistent across related lines and reduces the back-and-forth that usually fragments a session. North also records contextual ad-libs during the same session. The grunts, the breath sounds, the frustrated exhales after a missed shot. These are not added in post-production. They are captured live and mixed into the game audio at specific triggers. When Cassidy says "Not today" after a well-timed headshot, there is a separate audio file for the follow-up grunt that plays instantly. The timing is tight because the game engine reads the trigger frame and plays the sound within milliseconds. One practical issue I ran into during a modding project was trying to isolate and replace certain audio cues without breaking the sequencing. The Overwatch audio pipeline uses a custom implementation where some lines are prefetched based on predicted player actions. If you swap out a line without accounting for the prefetch buffer, the replacement audio desyncs by roughly two hundred milliseconds. The fix is to match the original file length exactly and keep the same silence padding at the start and end. I learned this the hard way after spending an afternoon debugging why my modified voice pack sounded sluggish during close-range engagements.
What Actually Happens In A Voice Session
A typical recording block for a hero like Cassidy runs about four to six hours with breaks. The actor arrives, warms up, and then reads through a script that is not organized by story but by audio trigger tags. Lines are grouped by emotional state and contextual category rather than narrative flow. You might record five lines about regret, then three about confidence, then two about humor, with no linear connection between them. The script itself is usually annotated with metadata. Each line has a tag for game context, a tag for emotional tone, and sometimes a tag for proximity. A whisper-quiet line recorded in a booth will sound different from the same line played through an in-game speaker model. The sound designers use convolution reverb to simulate the acoustic space of the map. This means the raw voice recording and the processed in-game version are two different things. North has mentioned in interviews that he sometimes records lines differently depending on whether the character is speaking to an enemy or to an ally, even when the words are the same. The intonation shifts subtly. The pitch drops slightly when addressing an opponent and lifts when addressing a teammate. These micro-adjustments are nearly imperceptible individually but create a layer of realism when the game triggers them dynamically.
Get the Full Details
The Technical Side Of Voice Implementation
Overwatch uses a proprietary audio middleware system built on top of standard game audio tools. Voice lines are not simply played when triggered. They go through a decision tree that evaluates distance, urgency, and recent audio history. If Cassidy just said "Draw your weapons," the game is less likely to trigger another opening line for several seconds. This prevents audio clutter and keeps the soundscape clean during chaotic team fights. The voice files themselves are stored as .ogg or proprietary formats depending on the platform. Each file contains multiple layers. A single line might have the main dialogue, a breath intake, and a lip-sync reference track bundled together. The lip-sync data is used by the animation system to drive facial bone weights in real time. Without it, the character mouth movements would not align with the audio, which breaks immersion noticeably at close camera distances. One counter-intuitive thing about this pipeline is that recording more lines does not always improve the character. The decision tree has a maximum activation rate per combat scenario. If you add too many similar lines, they compete for the same trigger window and the system randomly picks one, which can make the character sound repetitive rather than varied. Blizzard's audio team caps the line count per category to maintain diversity in what players actually hear during gameplay.
Common Misconceptions About The Recording Process
Many people think voice actors record all their lines in one day. For a hero with Cassidy's line count, that is impossible. The recording schedule spans multiple sessions over weeks. Each session focuses on a specific subset of lines, and the actor returns in different gear and mood states. The continuity is maintained through the script annotations and the director's notes rather than through a single continuous performance. Another misconception is that voice lines are recorded with full body movement and gesture tracking. They are not. North records standing or sitting in a soundproof booth with a microphone and a script on a stand. Some movement sounds are added later by Foley artists. The vocal performance is isolated from the physical performance, and the two are combined in post by the audio and animation teams separately. The rebrand from McCree to Cassidy did not require North to re-record the entire voice library. Most of the original lines were kept, with new lines added for the updated backstory and dialogue trees. The old McCree lines still play under certain conditions, which is why players who stuck with the game since launch sometimes hear overlapping references to the character's previous name in casual mode conversations.
Where To Find The Raw Audio Files
The official voice files are not publicly distributed by Blizzard. Any files you find online are extracted from the game installation directory using community tools. The extraction process involves unpacking the game's archive format, which changes with each patch. Tools like Overwolf's asset viewers or community-maintained repositories like Overwatch Assets on GitHub can pull the files, but the formats shift frequently enough that old extractors break after major updates. If you are looking for a specific line or want to analyze the recording quality, the extracted files are usually in .ogg format at 44.1kHz or 48kHz sample rate. Some lines include the dry vocal track, while others are already mixed with reverb and EQ. The dry tracks are harder to find because they are not always stored separately in the final build. You typically need to dig into the development builds or use community reverse-engineering efforts to access them. I have found that the most reliable source for current voice files is the community mirror maintained by long-time Overwatch modders. The repository gets updated within a few days of each patch, and the files are tagged with their trigger metadata so you can see exactly which in-game condition plays each line. This is more useful than raw audio dumps because it tells you not just what the line is but when and why it plays.
