upstream-archive/llama.cpp
1
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2026-08-06 10:00:48 -05:00
Code Issues Packages Projects Releases Wiki Activity

mtmd: correcting duplicate empty audio chunks for short inputs (#26536)

Browse Source 
* correcting duplicate empty audio chunks for short inputs

* tests.sh code restored
This commit is contained in:
Abhinay Krishna
2026-08-04 16:05:56 -04:00
committed by GitHub
parent a6aa6f5450
commit 474c92e722
1 changed files with 2 additions and 4 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
tools/mtmd/mtmd-audio.cpp
View File

@@ -556,10 +556,8 @@ bool mtmd_audio_preprocessor_whisper::preprocess(const float * s
}
std::vector<float> smpl;
// if input is too short, pad with zeros
// this is to avoid potential issues with stage1/2 padding in log_mel_spectrogram
// TODO: maybe handle this better
size_t min_samples = (size_t) hparams.audio_sample_rate * (hparams.audio_chunk_len + 1); // +1 second margin
// reflection padding needs one sample plus half an FFT window
size_t min_samples = (size_t) hparams.audio_n_fft / 2 + 1;
if (n_samples < min_samples) {
smpl.resize(min_samples, 0.0f);
std::memcpy(smpl.data(), samples, n_samples * sizeof(float));