Audio output has fixed ~+2.5 dB gain applied during render, causing clipping on normally-leveled sources

[endymuhardin]

Describe the bug

hyperframes render applies a constant ~+2.5 dB (ratio ~1.34×) gain to audio output, regardless of source configuration or data-volume setting. Sources with peaks already near -2 dBFS (common for voice recordings) are pushed past 0 dBFS, producing audible distortion.

Link to reproduction

https://github.com/endymuhardin/hyperframes-audio-gain-repro

Steps to reproduce

git clone https://github.com/endymuhardin/hyperframes-audio-gain-repro.git
cd hyperframes-audio-gain-repro
npm install
./generate-sample.sh   # deterministic 1 kHz sine, peak -6 dBFS
npx hyperframes render --quality draft --output renders/out.mp4
ffmpeg -i assets/sample.mp4 -af astats -f null - 2>&1 | grep "Peak level"
ffmpeg -i renders/out.mp4  -af astats -f null - 2>&1 | grep "Peak level"

Expected behavior

Output audio matches source levels (unity gain). Peak ≈ -5 dBFS in → peak ≈ -5 dBFS out.

Actual behavior

• Source peak: -5.12 dBFS → output peak: -2.19 dBFS (+2.93 dB)
• Source RMS: -9.05 dBFS → output RMS: -6.44 dBFS (+2.61 dB)
• Sample-by-sample cross-correlation against source shows an optimal scalar gain of 1.3517 (+2.62 dB)
• Residual after removing the gain is -47 dB below output — pure scalar multiplication, not a filter or mixed-in source
• The render also introduces a 1024-sample (~21 ms, one AAC frame) delay relative to source

A second reproduction against a voice recording (peak -2.17 dBFS source) measured gain 1.3325 (+2.49 dB) — magnitude consistent across content.

Workaround

data-volume="0.75" (≈ 1/1.3325) on the audio-bearing element restores unity output: measured gain 1.0020 (+0.02 dB). Requires the author to know about and manually compensate for a hidden boost.

data-volume="0" does not suppress the gain — audio still passes through at +2.5 dB.

Environment

Version          0.4.11 (latest)
Node.js          v25.9.0 (darwin arm64)
CPU              8 cores · Apple M1
FFmpeg           8.1
Chrome           system

Additional context

Reproducible both with single audio source (audioCount: 1, embedded video audio) and separate <audio> element (audioCount: 2). Gain is constant across both configurations. No mention of audio normalization/loudness in hyperframes docs topics or the installed skill references.