Skip to content

LLM inference in C/C++

C++ ggml

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

Abhinay Krishna474c92e722ce
diff --git a/tools/mtmd/mtmd-audio.cpp b/tools/mtmd/mtmd-audio.cpp
index 7fbc18ea93..ca4b64efa4 100644
--- a/tools/mtmd/mtmd-audio.cpp
+++ b/tools/mtmd/mtmd-audio.cpp
@@ -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));