Fix msvc bots build.

This started to happen after turning on "gn analyze" on trybots. It
looks like this code was never built on MSVC trybots.

This CL tries to avoid the type deduction.

Error:
quality_assessment/sound_level.cc(103):
    error C3535: cannot deduce type for 'const auto *' from '_FwdIt'
    with
      [
          _FwdIt=std::_Array_iterator<int16_t,1440>
      ]

Bug: webrtc:11262
Change-Id: Iea7cf2ec62f1d0edfcf6ceac169c92050339d3c4
Reviewed-on: https://webrtc-review.googlesource.com/c/src/+/172088
Reviewed-by: Per Ã…hgren <peah@webrtc.org>
Commit-Queue: Mirko Bonadei <mbonadei@webrtc.org>
Cr-Commit-Position: refs/heads/master@{#30933}
diff --git a/modules/audio_processing/test/py_quality_assessment/quality_assessment/sound_level.cc b/modules/audio_processing/test/py_quality_assessment/quality_assessment/sound_level.cc
index d58b57e..1f24d9d 100644
--- a/modules/audio_processing/test/py_quality_assessment/quality_assessment/sound_level.cc
+++ b/modules/audio_processing/test/py_quality_assessment/quality_assessment/sound_level.cc
@@ -100,9 +100,9 @@
     // Frame peak level.
     std::transform(samples.begin(), samples.begin() + audio_frame_length,
                    samples.begin(), [](int16_t s) { return std::abs(s); });
-    const auto* peak_level =
-        std::max_element(samples.begin(), samples.begin() + audio_frame_length);
-    const float level_curr = static_cast<float>(*peak_level) / 32768.f;
+    const int16_t peak_level = *std::max_element(
+        samples.cbegin(), samples.cbegin() + audio_frame_length);
+    const float level_curr = static_cast<float>(peak_level) / 32768.f;
 
     // Temporal smoothing.
     auto smooth = [&level_prev, &level_curr](float c) {