Encoding and decoding floats

[LatteSoyMilkCookie]

I have a stream of raw PCM samples stored as an array of floats that I want to encode then decode back into an array of floats.

The decoded audio does return an array of floats, but the values are all incredibly small, the first sample is -0.001953125 which turns into 3.0517578E-35, and it flipped from negative to positive! I understand OPUS is lossy, so I shouldn't expect the input and output to be exactly the same, but the decoded data cannot be turned back into pleasant audio like this.

Below is the code I'm using to encode and decode the array of floats:
float[] _currentData = new float[960] {PCM samples here};
byte[] encoded = new byte[1000];
int encodedBytes = encoder.Encode(_currentData, 960, encoded, encoded.Length);

float[] _decodedSamples = new float[960];
var decodedSamples = decoder.Decode(encoded , encodedBytes, _decodedSamples, 960, false);

Any help/advice is greatly appreciated, thank you!

[SineVector241]

I'm not sure what could be the issue.
Is your sample rate 48khz at 1 channel? And how long is the expected audio? 5ms, 10ms, 20ms, 40ms?

Also are both of your decoder and encoder the same samplerate and channels?

Also found a similar issue to yours. As suggested, try working with real audio samples like a sine wave and output it.
https://stackoverflow.com/questions/58204592/opus-encode-decode-no-error-but-not-the-same-value

[LatteSoyMilkCookie]

Thanks for the help! I am using real audio samples to test this, but I recorded it with 1 channel while the encoder and decoder were set for 2. The expected audio is 20ms and the sample rate of the decoder, encoder, and sample match. The data came out much better and I was able to use it to reconstruct a 1 second audio clip, but the first 108 float values for each decoded packet were still extremely small which led to the reconstruction sounding "robotic" Hopefully I'm just making some small mistake somewhere else and I'll get this working soon!

[SineVector241]

Yeh you'll need to match the number of channels from the original audio source for both the decoder and encoder.
Also opus has frame dependent packets so usually on the first few decodes, it might not be so accurate. I am using opus myself in my own project if you would like to take a look at how I do things but I use 16bit PCM decode/encode and convert to IEEEFloat for audio effect mixing.

https://github.com/AvionBlock/VoiceCraft/blob/dev/VoiceCraft.Client/VoiceCraft.Client/Network/VoiceCraftClient.cs#L144 - Encode

https://github.com/AvionBlock/VoiceCraft/blob/dev/VoiceCraft.Client/VoiceCraft.Client/Network/VoiceCraftClientEntity.cs#L151 - Decode

[LatteSoyMilkCookie]

Hey, thank you for all the help! I've finally got a p2p voice chat system running in unity because of this! OpusSharp is goated 🐐