From ac8f12f25994bcb66761aac9d5cb3c9defb0ebd6 Mon Sep 17 00:00:00 2001 From: Concedo <39025047+LostRuins@users.noreply.github.com> Date: Thu, 26 Feb 2026 17:50:49 +0800 Subject: [PATCH] still a bit wonky --- otherarch/acestep/dit-vae.cpp | 8 ++++---- 1 file changed, 4 insertions(+), 4 deletions(-) diff --git a/otherarch/acestep/dit-vae.cpp b/otherarch/acestep/dit-vae.cpp index a33c4b5c0..40b3a7f78 100644 --- a/otherarch/acestep/dit-vae.cpp +++ b/otherarch/acestep/dit-vae.cpp @@ -762,7 +762,7 @@ std::string acestep_generate_audio(const music_generation_inputs inputs) const char * keyscale = req.keyscale.empty() ? "N/A" : req.keyscale.c_str(); const char * timesig = req.timesignature.empty() ? "N/A" : req.timesignature.c_str(); const char * language = req.vocal_language.empty() ? "en" : req.vocal_language.c_str(); - float duration = req.duration > 0 ? req.duration : 60.0f; + float duration = req.duration > 0 ? req.duration : 120.0f; long long seed = req.seed; int num_steps = req.inference_steps > 0 ? req.inference_steps : 10; float guidance_scale = req.guidance_scale > 0 ? req.guidance_scale : 7.0f; @@ -868,8 +868,8 @@ std::string acestep_generate_audio(const music_generation_inputs inputs) // Context building // Silence latent for this T - // std::vector silence(Oc * T); - // memcpy(silence.data(), silence_full.data(), (size_t)(Oc * T) * sizeof(float)); + std::vector silence(Oc * T); + memcpy(silence.data(), silence_full.data(), (size_t)(Oc * T) * sizeof(float)); // Decode audio codes if provided int decoded_T = 0; @@ -895,7 +895,7 @@ std::string acestep_generate_audio(const music_generation_inputs inputs) for (int t = 0; t < T; t++) { const float * src = (t < decoded_T) ? decoded_latents.data() + t * Oc - : silence_full.data() + (t - decoded_T) * Oc; + : silence.data() + t * Oc; for (int c = 0; c < Oc; c++) context_single[t * ctx_ch + c] = src[c]; for (int c = 0; c < Oc; c++)