Are there better ways to apply audio gain and mixing?

[SmithGoll]

Describe the feature

The clipping issue in Sonivox is extremely severe.
The vast majority of cases occur during the final mixing and audio gain application stages.
It is possible that replace current mix and gain implement to prevent clipping and improve audio quality?

[rsp4jack]

Hi,
Could you please describe what "the final mixing and audio gain application stages" refer to?

About the clipping issue:
After EmbeddedSynth/sonivox-fork#80, the output of WT and FM synth, which is 16-bit full scale (1.0 = 2^16-1), is mixed into a 32-bit mixing buffer (still 16-bit full scale), and then mixed down to 16-bit output buffer.
Reducing the volume should help. But a probable better way is to output a 32-bit buffer with 16-bit full scale directly from the mixing buffer, or simply a float buffer, so that peaks over 1.0 are not clipped.

WT synth output 16-bit values to the mixing buffer because WT synth never apply gain over +0dBFS to the sample. FM synth outputs 32-bit values with 16-bit full scale, which is made in EmbeddedSynth/sonivox-fork#101.

[SmithGoll]

But a probable better way is to output a 32-bit buffer with 16-bit full scale directly from the mixing buffer.

It sounds good to prevent clipping, but can it really be achieved?

[rsp4jack]

An obstacle is the effect. Effects supports only EAS_PCM, which is 16-bit.

[pedrolcl]

I've also received a complaint on my Drumstick MIDI Player about Sonivox being very loud. I suppose that the user is referring to this clipping issue.

Seems that this depends on the use case: a player application using Sonivox is more susceptible to this problem because it typically will render songs using multiple channels that need to be mixed. On the other side, my virtual piano VMPK outputs a single channel each time, and I'm not receiving complaints being Sonivox the default/usual choice on Linux for this program.

One possible workaround is letting the user to adjust the gain parameter, like the sonivoxrender program does. The API function is EAS_SetVolume(). There is a bug in this function's header documentation, fixed in #29

sonivox/arm-wt-22k/host_src/eas.h

Lines 381 to 398 in bfbb0ce

	/*----------------------------------------------------------------------------
	* EAS_SetVolume()
	*----------------------------------------------------------------------------
	* Purpose:
	* Set the master volume for the mixer. The default volume setting is
	* 90 (-10 dB). The volume range is 0 to 100 in 1dB increments.
	*
	* Inputs:
	* pEASData - pointer to overall EAS data structure
	* volume - the desired master volume
	*
	* Outputs:
	*
	*
	* Side Effects:
	* overrides any previously set master volume from sysex
	*
	*----------------------------------------------------------------------------

If not specified, DEFAULT_VOLUME is 100, which means +0 dB, and the range is from 0 to 196. Depending on the usage of libSonivox, the client app may want to set a different default (for instance 90 if it is a SMF player) and also let the user to set the gain, like sonivoxrender does. This workaround is totally outside the Sonivox project, and depends only on the downstream projects.

What are your opinions?