Audio and Speech Processing Toolkits

Audio work adds a constraint that general DSP does not have: the result is judged by ear as well as by measurement. A change that looks small on a magnitude plot can be plainly audible, so the audio toolkits are built for auditioning while parameters change rather than for batch processing alone.

The chain is assembled from equalisation, dynamics and effect blocks. All of them are sample-rate agnostic and share one gain-staging convention, so inserting a block does not silently change the level reaching the blocks after it.

Equalisation and dynamics

Parametric equaliser sections are specified by centre frequency, gain and Q, and are implemented as cascaded biquads so a many-band curve stays numerically stable. Dynamics blocks work on threshold, ratio, attack and release, with the gain reduction plotted against time to make pumping and over-compression visible.

Speech and analysis

For speech, the frame-based view matters more than the global spectrum. Framing, pre-emphasis, spectral envelope estimation and pitch tracking operate on short overlapping frames, and the spectrogram is usually the most direct way to check the result.

Loudness and delivery

Delivery targets are stated in loudness units and true-peak level rather than sample peak. The metering block reports integrated, short-term and momentary loudness alongside true-peak, and the resampling stage converts to the delivery sample rate with a documented anti-alias filter.