diff options
| author | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2017-11-08 20:35:30 -0500 |
|---|---|---|
| committer | Jean-Marc Valin <jmvalin@jmvalin.ca> | 2017-11-09 22:45:32 -0500 |
| commit | ff98202e0a40cb4209ff85e2df3edcc3efc3d59f (patch) | |
| tree | 8bc5ae98d2456e6f77e46961472c9d9fb1dbc3b1 /src/analysis.c | |
| parent | 877d3d2cefb868aa0ebc1b000f16824b7bfeedac (diff) | |
| download | libopus-ff98202e0a40cb4209ff85e2df3edcc3efc3d59f.tar.gz | |
Retrained speech/music RNN
Larger network, using 8-bit weights
Diffstat (limited to 'src/analysis.c')
| -rw-r--r-- | src/analysis.c | 2 |
1 files changed, 1 insertions, 1 deletions
diff --git a/src/analysis.c b/src/analysis.c index 2e3913c9..11f87126 100644 --- a/src/analysis.c +++ b/src/analysis.c @@ -867,7 +867,7 @@ static void tonality_analysis(TonalityAnalysisState *tonal, const CELTMode *celt info->activity_probability = frame_probs[1]; /* It seems like the RNN tends to have a bias towards speech and this warping of the probabilities compensates for it. */ - info->music_prob = frame_probs[0] * (2 - frame_probs[0]); + info->music_prob = MAX16(1-10*(1-frame_probs[0]), MIN16(10*frame_probs[0], .12+.69*frame_probs[0]*(2-frame_probs[0]))); /*printf("%f %f %f\n", frame_probs[0], frame_probs[1], info->music_prob);*/ #ifdef MLP_TRAINING |