How do you optimize the trade-off between audio quality and file size?

Yes, for example , if I'm compressing music to fit more songs on a device, I might use lossy compression-mp3-wma-etc, which reduces the file size significantly but also slightly degrades the sound quality. By balancing the level of compression, we can achieve a file size that's manageable without making the audio quality noticeably worse.

Audio quality can be measured automatically using intrusive (reference required) metrics such as SI-SDR, SI-SNR, SIR, SAR and, specifically for speech, a range of perceptual metrics such as STOI or PESQ. All of these metrics have been found to be useful, but they don't always match human perception well. Others have attempted to use neural networks to assess audio quality, with works such as Noresqa, Visqol and DNMOS; this has also provided some interesting perspectives and has allowed evaluation without access to the reference audio. However, the true gold standard for assessing audio quality is the MUltiple Stimuli with Hidden Reference and Anchor (MUSHRA) method, but this involves humans in the loop.

When we are optimizing the trade-off between audio quality and file size, we focus on metrics like signal-to-noise ratio (SNR) or Total Harmonic Distortion (THD) or some ML metric such as RMSE and bit rate to make informed decisions. in compressing a podcast, we might choose a lower bit rate to keep the file size small while maintaining an acceptable SNR. This way, we ensure the audio is clear enough for listeners, without unnecessarily large files. for each different type of audio this threshold can be different.

if I want the smallest file size with decent quality, I usualy go with MP3 because it’s widely supported, even though it’s not the highest quality. If I need better sound quality and am okay with a slightly bigger file, AAC could be a better option, especially if the audio will be played on Apple devices. I think the key is to pick the standard that fits the specific needs , device and constraints of the task.

This can be very technical. If I want to use in my programming or easy in command-line (windows/linux/etc), I really like FFMPEG. also I usually use Adobe Audition or Medica Encoder. This software has many options and configuration.

Audio compression has become less important due to the reducing cost of communication and storage. As bandwidth and storage costs have decreased, it has become more cost-effective to store and transmit uncompressed audio than to compress it. This has resulted in a shift in focus from compression to quality. Advances in audio processing have made it possible to produce higher quality audio without the need for compression. Increasingly, audio engineers are focusing on improving the quality of sound rather than trying to squeeze more sounds into a smaller file size with compression. This has led to an increase in the use of lossless audio formats such as FLAC and ALAC, which allow for full-resolution audio playback without any loss in quality

Human ears do not have uniform sensitivity across their hearing ranges, so objective measurements like total SNR may be misleading. Using an A-weighted or otherwise frequency-dependent SNR function would be more appropriate, as it would more accurately reflect the human perception in quality.

The increased focus on quality over compression has also had an impact on streaming services. As streaming services become more popular, they are placing a greater emphasis on providing high-quality audio to their users. This has resulted in the adoption of streaming formats such as Dolby Atmos and DTS:X, which provide users with high-resolution audio without any compression. Audio compression has become less important due to the reducing cost of communication and storage, and the increasing focus on providing high-quality audio to users.