mtmd: refactor audio preprocessing

[ngxson]

The goal of this PR is to allow more audio pre-processing mechanism to be added into mtmd

While the code is not very clean, this should already allow:

• Simplify model : add ASR support for LFM2-Audio-1.5B #17694
• Potentially support gemma 3n audio

---

Key points

• Each model's preprocessor now have their own subclass extended from mtmd_audio_preprocessor
• Preprocessor can access hparams directly (to read audio params like n_mel, n_fft, etc)
• Each preprocessor also have its own initialize() function which will be called on model load, to initialize global cache entries like sin/cos, hann window
• Filter bank is now constructed dynamically thanks to @tdakhran 's implementation of fill_mel_filterbank_matrix (the hard-coded value is now removed)

[ngxson]

Hmm, I think I can also upstream some changes from #17694 , that would make your PR a bit shorter @tdakhran

I will remove the pre-calculated filters and replace with your version

Edit: since my goal is to implement conformer, I think I will end up copying a lot of code and refactor them along the way

[ngxson]

@ggerganov This is ready for review. I only have basic knowledge about signal/audio processing, would appreciate if you can have a deeper look to see if things are still correct compared to the original code from whisper.cpp

Note: this change also contain enough code for LFM2-audio and gemma 3n audio preprocessor

Test results:

[audio]  OK:   ggml-org/ultravox-v0_5-llama-3_2-1b-GGUF:Q8_0
[audio]  OK:   ggml-org/Qwen2.5-Omni-3B-GGUF:Q4_K_M
[audio]  OK:   ggml-org/Voxtral-Mini-3B-2507-GGUF:Q4_K_M

[ggerganov]

I didn't spot anything suspicious. Though ultimately we have to do some actual tests to be sure that the Whisper still works.

I assume @tdakhran did the necessary verification, so I think it's OK to proceed.

[tdakhran]

@ngxson, thanks for the refactor.

@ggerganov I've verified that generated coefficients match existing for n_fft = 400.

[tdakhran]

params.use_natural_log = true; for LFM2-Audio-1.5B, I'd like to avoid reimplementing the whole processor just because it. Shall all params members be defined in hparams?

[ngxson]

For other models, it's recommended to make a dedicated class that extends from mtmd_audio_preprocessor:

struct mtmd_audio_preprocessor_lfm2a : mtmd_audio_preprocessor {
    mtmd_audio_preprocessor_lfm2a(const clip_ctx * ctx) : mtmd_audio_preprocessor(ctx) {}
    void initialize() override;
    bool preprocess(const float * samples, size_t n_samples, std::vector<mtmd_audio_mel> & output) override;
};

This way, you can customize initialization of cache, while also allow defining custom filter params and handling custom paddings

[tdakhran]

sounds good, will create a dedicated class.

[tdakhran]

@ngxson for reference this is how rebase LFM2 Audio ASR PR ngxson#58 look