opus - Jun 2018 - OPUS on cortex M4

Hi everybody

I wrote a couple of months ago about implementing the opus codec on a DK52 from
Nordic (nrf52832 MCU, Cortex M4). In the meantime I got a bit further and would
again like to ask some questions.

My settings:
I included all the opus files into my project (opus-1.2.1)
Defined: FIXED POINT, VAR_ARRAYS, DISABLE FLOAT API, OPUS_BUILD, CELT_MODE_ONLY

I am able to allocate enough memory for the encoder as well as decoder state and
they both return with OPUS_OK.

    uint32_t error;
    opus_int32 Fs = 8000;
    int channels = 1;

    OpusEncoder *enc;
    enc = opus_encoder_create(Fs, channels,
OPUS_APPLICATION_RESTRICTED_LOWDELAY, &error);

    OpusDecoder *dec;
    dec = opus_decoder_create(Fs, channels, &error);

However, If I try to make a simple example of opus_encode followed by and
immediate opus_decode I do not get reasonable values compared to the input.
I'm wondering if I'm somehow misunderstanding how the codec works.
Could somebody tell me if my approach could work?

    int frame_size = 160;
    opus_int16 test_data_size = frame_size*channels*sizeof(opus_int16);
    opus_int32 max_data_bytes = 100;

    opus_int16 test_data[test_data_size];
    opus_int16 *p_test_data = test_data;
    unsigned char encoded_buffer[max_data_bytes];
    unsigned char *p_encoded_buffer = encoded_buffer;
    opus_int16 decoded_buffer[test_data_size];
    opus_int16 *p_decoded_buffer = decoded_buffer;

    for(uint16_t i = 0; i < test_data_size; i++)
    {
    test_data[i] = 100;
    }

    opus_int32 encoded_size = opus_encode(enc, p_test_data, test_data_size,
p_encoded_buffer, max_data_bytes);
    NRF_LOG_INFO("encoded size is %d", encoded_size);

    opus_int32 decoded_size = opus_decode(dec, p_encoded_buffer, encoded_size,
p_decoded_buffer, test_data_size, 0);
    NRF_LOG_INFO("decoded size is %d", decoded_size);

Opus_encode returns 55 as encoded length, which seems reasonable to me.
Opus_decode returns 320.
The decoded values start out at 0, go to -16 up, then up to 135 and then slowly
down to the 17 (see attached file).
Furthermore, I do not see much of a change if I increase the complexity.

I really think that I'm close to getting it working, but I'm definitely
still missing something.
I would really be glad for any input.

Best,
Nick



-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.xiph.org/pipermail/opus/attachments/20180629/5a28d662/attachment-0001.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: opus_sample_results.pdf
Type: application/pdf
Size: 47381 bytes
Desc: opus_sample_results.pdf
URL:
<http://lists.xiph.org/pipermail/opus/attachments/20180629/5a28d662/attachment-0001.pdf>

Ehrenberg  Nicolas wrote:
>      for(uint16_t i = 0; i < test_data_size; i++)
>      {
>      test_data[i] = 100;
>      }
>
> [snip]
>
> The decoded values start out at 0, go to -16 up, then up to 135 and then 
> slowly down to the 17 (see attached file).
> Furthermore, I do not see much of a change if I increase the complexity.
You're encoding a signal with a constant (DC) offset. opus_encode() 
includes a 3 Hz highpass filter to remove the (inaudible) DC component, 
because it screws up other parts of the codec (mainly it would create 
MDCT leakage that is completely unmasked). So you shouldn't expect to 
get your constant values of 100 back.

In general the numeric values of the output aren't guaranteed to 
resemble the numeric values of the input. The codec will just try to 
make the two signals sound the same to human ears.