timing report for wald_asr, using speech command dataset

closes #5 (closed) (if everyone is cool with time.perf_counter)

• removed all tiere dataset related code
• ported over speech commands dataset code
• wrote a function that generates a timing report for wald_asr:
  • reading files from disk
  • feature extraction
  • training (obviously this won't run on the device itself, but I included it for now)
  • prediction

---

on the pi:

2000 files
3 rounds
15 segments
15 cepstra
250 estimators

--- reading 2000 files ---
              method      avg  per-file
0  time.perf_counter  5.74513  0.002873

--- feature extraction ---
              method       avg  per-file
0  time.perf_counter  91.32158  0.045661

--- training ---
              method       avg  per-file
0  time.perf_counter  98.99061  0.049495

--- prediction ---
              method      avg  per-file
0  time.perf_counter  2.47994   0.00124