Benchmarks use the music.wav file, we use *divan* as benchmark harnass.
The time needed to load the wav file is excluded from the benchmark
by preparing the data into a special test Source. That source also
enables converting between formats.
In the future *divan* will add support for structured (json) output.
Then we could integrate with the bencher service to generate benchmark
reports for all PR's and keep a timeseries of performance.
