Used "Feature" Issue Type as closest I could find to "Further Work".
When working on #4, it was noted that when the voice.wav sample was encoded/decoded using the RADE pipeline some artifacts were present in the decoded speech for both RADE V1 & V2. This is most obvious through ear bud type headphones (better high freq response) and sounds like aliasing. The decoded speech sounds better through laptop and 50mm external speakers. The input sample does have some background hiss and (possibly) some acoustic echo/reverb.
The sample loss is fine, around 0.093 (V1) and 0.064 (V2) when encoded with either the Python inference.sh too or the rade_c wav tx/rx tools (--start_clip 50 --end_clip 50 for the C version to avoid start/stop transients).
It would be worth exploring why this artifact occurred as further work.
The sample has been replaced in the rade_c but can be recovered using:
git show 7dd041e:voice.wav > voice_original.wav
Used "Feature" Issue Type as closest I could find to "Further Work".
When working on #4, it was noted that when the
voice.wavsample was encoded/decoded using the RADE pipeline some artifacts were present in the decoded speech for both RADE V1 & V2. This is most obvious through ear bud type headphones (better high freq response) and sounds like aliasing. The decoded speech sounds better through laptop and 50mm external speakers. The input sample does have some background hiss and (possibly) some acoustic echo/reverb.The sample
lossis fine, around 0.093 (V1) and 0.064 (V2) when encoded with either the Pythoninference.shtoo or therade_cwav tx/rx tools (--start_clip 50 --end_clip 50 for the C version to avoid start/stop transients).It would be worth exploring why this artifact occurred as further work.
The sample has been replaced in the
rade_cbut can be recovered using: