Audio Latency in Film Production: Recording, Playback, and Post Sync Problems Solved

The Playback Delay That Made the Actor Mistime Every Take

A director is shooting a dialogue scene that requires an actor to respond to pre-recorded audio playback -- a radio broadcast the character hears through a prop speaker. The playback is routed through the sound department's laptop and a small reference speaker. The actor rehearses the timing. When cameras roll, the audio playback reaches the actor's ears approximately 85 milliseconds late -- barely perceptible as a discrete delay, but enough to shift the actor's rhythm in every take.

Nobody identifies the problem on set. The delay isn't obvious on the scratch audio monitoring. It's only in the edit, when the picture cut of the actor's responses is assembled against the final mixed audio track, that the rhythm of the scene feels subtly wrong. The actor is reacting to something that, on the final cut, sounds like it happened before the response.

The cause was a combination of the laptop's USB audio interface running at a 512-sample buffer (approximately 11ms at 48kHz) and the Bluetooth monitoring speaker adding another 70ms of Bluetooth encoding latency. The total round-trip latency from digital audio to the actor's ears was 81ms -- well above the 25ms threshold where latency becomes perceptible to performers in a synchronous task.

Audio latency is invisible until it causes a problem. This post explains where it comes from, how to calculate it, and how to eliminate it in the critical monitoring and recording paths.

Technical latency figures in this post are drawn from Audio Engineering Society (AES) papers on monitoring latency perception, Universal Audio's published buffer latency data, and Sound Devices' MixPre series technical specifications.

What Audio Latency Is and Where It Comes From

Latency in a digital audio system is the time delay between a sound entering the system (a microphone capturing audio, a playback command being issued) and that sound being heard or recorded at the output. It accumulates at every stage of the signal chain.

Analog-to-digital conversion (ADC latency) occurs when a microphone signal is converted from an analog electrical signal to digital samples. At 48kHz, one sample represents approximately 0.021 milliseconds. ADC converters in professional audio interfaces add approximately 1-3ms of inherent conversion latency regardless of buffer settings.

Buffer latency is the largest source of DAW monitoring latency. A digital audio interface collects incoming samples into a buffer before processing them. The buffer size is set in the DAW's audio preferences -- typically ranging from 32 to 2048 samples. Latency in milliseconds = buffer size in samples / sample rate in Hz x 1000. At 48kHz with a 512-sample buffer: 512 / 48000 x 1000 = 10.67ms. With a 2048-sample buffer: 42.67ms. For recording sessions, lower buffer sizes reduce latency but increase CPU load. For mixing sessions where real-time monitoring is less critical, larger buffers reduce processing glitches.

Plug-in latency accumulates when DAW plug-ins that require look-ahead processing (certain compressors, limiters, spectral processors) introduce a fixed delay in their processing chain. DAWs with PDC (Plugin Delay Compensation) compensate for this automatically, but PDC only compensates within the DAW -- it doesn't reduce the latency the performer hears in a recording monitoring path.

Digital-to-analog conversion (DAC latency) mirrors ADC latency on the output side -- another 1-3ms depending on the converter.

Speaker or headphone latency varies dramatically: a wired speaker connected to an audio interface output has near-zero additional latency after the DAC. Bluetooth headphones and speakers add 30-200ms of encoding and transmission latency depending on the Bluetooth codec (aptX Low Latency achieves approximately 32ms; standard SBC Bluetooth adds up to 250ms). HDMI audio output through a television or monitor adds 50-200ms depending on the display's internal processing.

Total Round-Trip Latency Reference

The threshold at which monitoring latency becomes perceptible to most performers is approximately 20-25ms for rhythmic tasks (playing to a click track) and approximately 30-35ms for speech/dialogue. Below 10ms is considered transparent for essentially all applications.

Three Real-World Latency Scenarios

Example 1: Location Sound Recording, Monitoring Latency on Set

A production sound mixer on a 10-day feature shoot using a Sound Devices 688 recorder and a pair of Sennheiser MKH 50 boom microphones. The director of photography wants to monitor the location audio through the recorder's headphone output to check for background noise between takes.

Latency concern: The Sound Devices 688 uses direct-through monitoring -- the headphone output taps the analog signal before ADC conversion. Monitoring latency is essentially zero: the DP hears the microphone signal in real time with no digital processing delay.

Scenario where this matters: When the production switched to using the recorder's USB output to send audio to a MacBook for recording a scratch audio backup, the MacBook's Core Audio system added 256-sample buffer latency (approximately 5.3ms at 48kHz) plus USB protocol overhead (approximately 2ms). Total: approximately 7ms. Acceptable for the scratch backup use case, but the sound mixer correctly kept their primary monitoring on the 688's direct output rather than the MacBook's return signal.

Key decision: Direct analog monitoring paths (built into most professional location recorders) bypass digital latency entirely. Plug-ins or digital processing on the monitoring path introduce latency. Keep the monitoring path as short and analog as possible on set.

Example 2: Post-Production Session, Dialogue Editing with Plug-In Latency

An editor cutting dialogue in DaVinci Resolve Fairlight on a Mac Studio (M2 Max). The session includes FairlightFX noise reduction, iZotope RX7 Connect, and a Waves DeEsser on every dialogue track. The DeEsser has a 256-sample lookahead, the RX7 module adds 512 samples of latency, and FairlightFX noise reduction adds 128 samples.

Problem: With PDC enabled in Resolve, all tracks are automatically delayed to compensate for the plug-in with the highest latency (the RX7's 512 samples = 10.67ms at 48kHz). The session plays back correctly in sync. However, when the editor exports a rough cut and plays it in QuickTime Player, the audio is 10.67ms late relative to the video because QuickTime is playing the exported file without the PDC offset.

Diagnosis using the [Audio Latency Calculator](/tools/audio-latency): Entering the RX7's latency (512 samples at 48kHz), the DeEsser (256 samples), and FairlightFX (128 samples) confirms the total session latency is 512 samples = 10.67ms. At 24fps, 10.67ms represents approximately 0.26 frames -- sub-frame and not visible on close inspection.

Resolution: The editor verified that the final export from Resolve was rendering with compensated audio (Resolve includes latency in the render, unlike real-time monitoring). The QuickTime player discrepancy was confirmed to be a monitoring-only issue -- the exported file was in sync. No action required beyond confirming the export path handled PDC correctly.

Example 3: Music Recording Session, Performer Monitoring Latency

A composer recording live violin for a film score on a Mac Pro with Pro Tools Ultimate and a UAD Apollo X16 interface. The violinist monitors through the Apollo's hardware monitoring path (Realtime Analog Classics plug-ins run on the Apollo's FPGA processors at near-zero latency) via closed-back headphones.

Configuration: Apollo hardware monitoring at 64-sample buffer = 1.3ms total round-trip latency. The violinist hears their own playing with approximately 1.3ms of delay -- effectively transparent.

The problem that occurred: A session engineer accidentally switched the violinist's monitoring from the Apollo's hardware monitoring path to Pro Tools' software monitoring path. With a 64-sample buffer in Pro Tools, software monitoring latency is 64/192000 x 2 (round-trip) + USB overhead = approximately 1ms additional -- still acceptable. But the engineer had the session running at 192kHz for hi-res delivery, and a 64-sample buffer at 192kHz = 0.33ms per buffer, but Pro Tools' software monitoring with real-time plug-ins (a reverb was active in the monitoring chain) at 192kHz added 6ms of additional processing latency. The violinist began playing sharp -- a classic symptom of monitoring latency causing pitch correction in a performing musician.

Resolution: Switched back to hardware monitoring on the Apollo. The Apollo's FPGA processors run independently of the Mac's buffer size at effectively zero latency regardless of DAW buffer settings.

Diagnosing and Fixing Audio Latency: A Troubleshooting Workflow

Step 1: Identify which part of the chain has the latency. The Audio Latency Calculator accepts buffer size, sample rate, and optional plug-in latency values to calculate total system latency. Work through the signal path: ADC latency (check manufacturer spec for your interface), buffer latency (read from DAW audio preferences), plug-in latency (available in plug-in documentation or the DAW's mixer plug-in info display), DAC latency, and output device latency.

Step 2: For recording monitoring latency, use direct hardware monitoring if available. Most professional audio interfaces -- Apollo, RME, Universal Audio, Audient -- include a hardware monitoring path that bypasses the DAW's software processing entirely. The direct monitoring path routes the microphone signal to the headphone output before ADC conversion, producing near-zero latency. Enable hardware monitoring for recording sessions, disable it for mixing sessions where plug-in processing on the monitoring chain is desirable.

Step 3: Reduce buffer size for recording sessions, increase for mixing sessions. For sessions where performers need low-latency monitoring, use the lowest buffer size your CPU can sustain without audio dropout. On an Apple M2 Max, a 64-sample buffer at 48kHz is typically stable for up to 40 tracks with standard plug-in loads. On an older Intel Mac, 256 samples may be the stable floor. Test by playing the session with all tracks active and gradually reducing buffer size until drop-outs appear, then set one step higher.

Step 4: Remove Bluetooth from any monitoring path used in synchronous performance. Replace Bluetooth speakers and headphones with wired equivalents for any recording, playback-to-picture, or actor-monitoring scenario. The 30-200ms of Bluetooth latency is incompatible with frame-accurate production audio and will cause rhythm and timing problems in any synchronous performance task.

Step 5: Check HDMI monitor audio latency when playing back to picture on set. If the director or DP monitors a video village feed with the audio output through an HDMI television or consumer monitor, the display's internal processing adds 50-200ms of latency. This is a monitoring-only problem -- it doesn't affect the recorded files. But it means the director hears audio out of sync with picture, which can produce confusing feedback about a scene's rhythm. Use a dedicated video monitor with simultaneous audio output via the mixer's headphone or speaker output, bypassing the HDMI audio path.

Step 6: Verify final exports for sync accuracy using a sync test tone. Before delivering a finished mix, generate a 1kHz test tone at frame 01:00:00:00 in the timeline. Export the final mix and re-import it into a new empty session. Confirm that the test tone aligns to 01:00:00:00 in the re-imported file. Any offset indicates that the DAW's export path is applying a latency offset that is not being compensated in the bounce.

Pro Tips and Common Mistakes

Pro Tip: For dialogue recording on location where the sound mixer also needs to feed ISO tracks to a mixer-recorder while monitoring through headphones, verify that the recorder's headphone monitoring path is tapping the analog mix bus directly, not the digital return. Sound Devices and Zaxcom recorders are designed this way by default. Some lower-cost recorders route the headphone output through the digital processing chain, adding buffer latency to the monitoring path and making it harder to judge sync between a boom mic and a second camera's scratch audio.

Pro Tip: When setting up a video village monitoring system for picture plus audio, use a dedicated reference monitor with an audio line-level input rather than the HDMI audio output of the video monitor. Connect the audio feed from the sound cart directly to the reference speaker's line input. This eliminates HDMI processing latency from the director's monitoring path and gives them a frame-accurate audio reference for evaluating takes.

Pro Tip: For post-production sessions where latency is causing visible sync issues in playback but the cause is unclear, use DaVinci Resolve's Sync Bin alignment tool or Pro Tools' Elastic Audio auto-analysis to measure the actual audio-to-picture offset. A consistent offset of 2-5 frames across all clips usually points to a buffer or interface latency issue that's been present since recording. An offset that varies per clip usually indicates timecode or sample rate mismatch rather than latency.

Common Mistake: Using the DAW's plug-in delay compensation value as the only guide to total system latency. PDC compensates for plug-in latency within the playback engine -- it does not reduce the latency a performer hears when monitoring through the software path. A session with 30ms of PDC-compensated latency still produces 30ms of monitoring delay for the performer. PDC is for mixing accuracy; hardware monitoring is for recording latency reduction.

Common Mistake: Assuming that a "low latency mode" in a DAW eliminates all latency. Low latency modes (available in Pro Tools, Logic Pro, and Cubase) temporarily bypass high-latency plug-ins to reduce monitoring latency during recording. They do not bypass buffer latency or interface conversion latency. At 256 samples / 48kHz, even in low latency mode, the remaining system latency is approximately 10ms -- acceptable for most recording situations but worth verifying against the threshold for your specific use case.

Frequently Asked Questions

What buffer size should I use for recording a film score with live musicians?

For live musicians monitoring through the DAW's software path, use the lowest buffer size your system can sustain without dropouts -- typically 64 or 128 samples on a modern computer. At 48kHz, 64 samples produces approximately 2.7ms of total round-trip monitoring latency, which is below the perception threshold for all musical tasks. If your system cannot sustain 64 samples with your full plug-in load, use hardware monitoring on your audio interface (Apollo, RME, MOTU) which bypasses DAW buffer latency entirely, and increase the buffer to 512 samples for stable processing.

How does audio latency affect sync when recording dialogue on set?

On a properly configured location sound setup with a professional recorder and wired boom microphones, audio latency during recording is negligible -- the recorder's ADC adds approximately 1-2ms of conversion latency, which is sub-frame at 24fps. The sync problem arises in monitoring: if the director or script supervisor is monitoring the audio through a system with significant latency (Bluetooth speakers, laptop speakers, HDMI monitors), they hear the audio late relative to picture. This affects their ability to judge the sync of a performance in real time. The recorded files themselves are accurately timed; the monitoring environment creates a false impression of sync drift.

Why does my exported mix sound out of sync when I import it back into my DAW?

The most common cause is that the DAW exported the audio with plug-in delay compensation applied but positioned the audio at the uncorrected timeline position. The result is a file that is internally correct but starts N samples later or earlier than expected. Check your DAW's export settings for an "offset" or "start timecode" option. In Pro Tools, exporting from the track's region start handles PDC correctly; bouncing to disk with a complex routing can sometimes produce an offset. Verify every final export with the sync test tone method described in the step-by-step section above.

Is audio latency a concern in DCP theatrical playback?

No. DCP servers (Barco Alchemy, Doremi, QUBE) synchronize audio and video from a single encrypted container file -- there's no separate audio interface or buffer path that can introduce differential latency. The ADC and DAC latencies in a theatrical system are designed and calibrated to maintain frame-accurate sync within the DCP playback engine. Audio latency is a concern during production and post-production mixing; it's not a variable factor in theatrical exhibition.

How do I know if my audio interface is introducing more latency than advertised?

Measure it using a loopback test: connect the interface's output directly to its input, generate a test click in the DAW, and measure the time between the generated click and its appearance in the recorded return signal. The elapsed time is the actual round-trip latency of the interface at the current buffer setting. Compare this to the manufacturer's stated latency specification. Most professional interfaces match their specs closely. Consumer-grade USB interfaces and built-in laptop audio occasionally add undocumented OS-level latency that exceeds the hardware spec, particularly on Windows systems without ASIO drivers.

Related Tools

The Audio Latency Calculator accepts buffer size, sample rate, and plug-in latency parameters to calculate total round-trip latency and convert it to milliseconds and frame equivalents at standard frame rates. For understanding how audio sync integrates with the post-production workflow beyond latency, timecode in film production covers the frame-addressing system that keeps picture and sound aligned across the entire post chain. The post on audio delivery standards for film and television covers the output side -- the platform specs your final mix must meet after you've resolved any latency issues in the production chain. The Sample Rate Converter is useful when latency investigations reveal a sample rate mismatch between recording and delivery formats.

Conclusion

Audio latency is a systemic property of digital audio chains. It accumulates at every stage -- conversion, buffering, processing, output -- and its effects range from transparent to catastrophic depending on where it occurs and what depends on accurate real-time audio response. The fix is almost always the same: shorten the monitoring path, eliminate Bluetooth, use hardware monitoring for recording, and verify exports against a sync reference tone.

This post covers standard stereo and multi-track audio chains. Networked audio systems (Dante, RAVENNA, AES67) introduce network-specific latency variables that require their own analysis tools and fall outside the scope of a production monitoring guide.

What's the largest latency you've identified in a production monitoring chain -- and was it visible on picture before you measured it numerically?