I’m doing a program that analyses music, but I’m having a problem with FSOUND_Sample_Load. Why does it have to load the entire song at once? Why can’t it load and decompress only the samples that I lock with FSOUND_Sample_Lock?
When I load a 60 minute mp3, it takes some 600 MB of RAM…
I had the crazy idea of splitting the mp3 into X parts (perhaps each part should be 1 MB uncompressed), and then process the parts. However, I don’t quite know how to split it correctly… I split it at a even 417 byte boundary (each mp3 frame should be 417 bytes, right?), but given such parts, FSOUND_Sample_Load sometimes complains about it not being correct… And I still haven’t figured out how to deal with ogg or wma files…
Does anyone have an idea on how to achieve this?
- kittelmann asked 14 years ago
No, I just realized that using a DSP is a bad solution, since a DSP is supplied with sound at at most some 700kSamples per seconds, which means that it’s a much slower solution than using sample_lock (which runs at an equivalent of some 3MSamples/second at my comp).
I guess I have to go for manual splitting of the files, but it’s a pain. Too bad fmod doesn’t have a solution for this.
What I do now is to split the mp3 and load it X frames at a time. I pick out X frames and then append a ID3 tag (is it neccessary to add the tag?). Then I feed that block of memory to FSOUND_Sample_Load. However, what is really interesting is that I get Y samples out, but about 25% of the times I call FSOUND_Sample_Load I get Y/2 samples 😮 . Excatly. I have checked that the frame header (incl. sync, stereo bits, bitrate, and so on) are identical every time I call FSOUND_Sample_Load.
Does anyone have a clue to what the problem might be? Is there a problem calling FSOUND_Sample_Load on the same index multiple times?
But if I have a mp3 which when loaded as one piece contains 10Msamples, and when loaded as 2 equal sized pieces (half the length of the original mp3) contain for instance 5M samples + 2.5 Msamples; is it not doing something wrong then 👿 .
All headers in the loaded frames are the same (appart from pad), and the memLength fed to FSOUND_Sample_Load are the same. FSOUND_GetError returns 0 after all calls.
I can accept that the mp3 decoding isn’t as perfect when loading the data this way, but I don’t want to loose half of the samples when loading parts of the song.
I had this new idea of using a stream instead (since that is lazy loaded), and then use FSOUND_Stream_GetSample from a DSP to get the samples. If I then put the frequency as high as possible, and then return nothing but silence from the DSP, then perhaps it could be possible.
What do you think about that?
Please login first to submit.