A compact, quantized runtime with built-in control a practical basis for on-device semantic audio in Internet-of-Sounds settings and a case study of the steering interface generates music carrying taste associations with genuine but bounded control for a subset of attributes.
Abstract
Semantic audio applications increasingly require controllable generation on commodity and embedded hardware rather than through framework-heavy datacenter stacks. We present \textit{aria}, a dependency-free native runtime that runs the complete text-to-music pipeline of Stable Audio~3 (SA3) on ordinary GPUs, CPU-only machines, and a Raspberry~Pi~5, with no Python or deep-learning framework underneath. Our main contribution is a study of quantization: running the model at lower numerical precision to fit tight memory budgets, saving memory in place rather than adding to it. Because the runtime owns every internal tensor, it also exposes activation steering, a low-cost way to steer what the model generates. We judge the quality cost with three independent measures of the output (prompt adherence, overall audio quality, taste preservation), each compared against the ordinary variation between random seeds. Eight-bit precision shows no measurable quality loss on any measure while sharply cutting memory, and it is the fastest mode on the GPU; four-bit adds a small, bounded cost but shrinks the footprint enough to run the $1.2$-billion-parameter model on an $8$\,GB Pi. Against the official implementation, aria matches or exceeds generation speed and starts about seven times faster. A case study of the steering interface generates music carrying taste associations (\emph{sonic seasoning}), with genuine but bounded control for a subset of attributes. These results make a compact, quantized runtime with built-in control a practical basis for on-device semantic audio in Internet-of-Sounds settings. The \textit{aria} runtime is released at https://github.com/matteospanio/aria.
The two studies show that closed-loop GPU fine-tuning does not guarantee monotonic mobile gains, especially on harder classification tasks, and that multi-dataset, on-device measurement is needed to stress-test deployment objectives.
Saif U Din, Muhammad Hussain, R. Timofte et al.· 0 citations
FluxBin is proposed, an algorithm-kernel co-design that synergizes post-training quantization with a highly optimized CUDA kernel and introduces Decoupled Row-Column Binary Decomposition to enhance representational capacity while maintaining hardware efficiency.
Qingyao Yang, Runming Yang, He Xiao et al.· 0 citations
Stable 4-bit floating-point (FP4) pretraining is difficult because the E2M1 payload represents only a narrow range of magnitudes. NVIDIA's Transformer Engine \nv{} recipe addresses this with current-tensor scaling, a randomized Hadamard transform (RHT), and bfloat16 (BF16) final layers, adding work outside the FP4 matrix multiplications. We instead pair E2M1 payloads with unsigned E5M3 (\ue{}) block scales. Their wider range permits periodic tensor scaling, while our recipe applies selective stochastic rounding to backward gradients, omits RHT, and uses FP4 in all eligible internal linears. We pretrain a Nemotron-H 8B model for nearly 190 billion tokens. Compared with Transformer Engine \nv{}, the proposed block-16 recipe finishes with lower final-window training loss and, under their respective quantized-inference policies, lower validation loss measured as held-out negative log-likelihood. Its quantized-inference downstream point estimates are also higher on all three reported aggregates. A native \nv{} execution ablation that jointly removes RHT and the BF16 final-block exemption increases measured model-body token throughput by 21.2\%. These results demonstrate end-to-end software-emulated \uefp{} pretraining with a simpler recipe and motivate native support for \ue{} block scaling.
Open-source Speech-to-Text (STT) systems are increasingly deployed across a wide range of hardware, from GPU servers to resource-constrained edge devices. However, these systems are rarely evaluated jointly across heterogeneous platforms with respect to accuracy, latency, and resource usage, making deployment decisions difficult in practice.We present a cross-hardware benchmark of six open-source STT engines (Whisper, whisper.cpp, Sherpa-ONNX, Silero, Vosk, Pocketsphinx) across four platforms—Linux GPU server, macOS, Windows, and Raspberry Pi 5. Using a unified and reproducible evaluation harness, we measure accuracy (WER/CER), latency and throughput, and CPU/GPU/memory usage on clean, noisy, and multilingual datasets at two scales (1.5k and 10k audio clips).Our results show that STT performance is deployment-dependent rather than universal. Sherpa-ONNX yields the lowest error rates on desktop platforms, Silero provides the lowest CPU latency for interactive use, and Whisper delivers the highest throughput under GPU acceleration. On Raspberry Pi 5, only lightweight models achieve near-interactive performance, exposing a fundamental accuracy–latency trade-off in edge environments. We further observe that robustness to noise does not necessarily correlate with clean-speech accuracy, highlighting the importance of evaluation under realistic conditions.This work provides actionable deployment guidance and a reproducible benchmark that surfaces system-level trade-offs in open-source STT across the edge–cloud continuum. Scripts, model configurations, and environment manifests are provided to support replication and extension on new hardware.
Uyen Thy Duong, Pakeeza Akram· International Conference on...· 0 citations
ExactMoE, an inference design that applies symmetric group-128 four-bit weight quantization only to routed experts, stores those experts in kernel-native MARLIN form in pinned host memory, and executes all selected experts through a configurable GPU-resident slot cache and fused grouped MoE kernels, identifies a practical memory-transfer-throughput frontier for complete-expert MoE inference.
This is an implementation and measurement study of what it costs to run a streaming speech enhancer on a CPU. We port FastEnhancer-Medium at 48 kHz to faster-enhancer.c, a C runtime with six int8 GEMM tiers selected at initialization, leaving architecture and weights untouched. One Apple M2 core reaches 0.069 real-time factor, against 0.230 for the fp32 ONNX Runtime graph on the same machine, a 3.3x speedup. A Galaxy S23+ (Snapdragon 8 Gen 2) reaches 0.096. The speedup comes from specializing every layer of the runtime around one fixed model. Activation ranges are recomputed per frame, so no calibration set is needed; the k=3 convolutions use Winograd F(2,3); cross-stage state is fp16; the GRU and the dequantization epilogues are fused; and nothing is allocated after startup. Over 824 VoiceBank-DEMAND utterances the engine tracks fp32 to within -0.006 PESQ and -0.08 dB SNR. Speed alone does not settle deployment cost. The enhancer holds a fraction of a core for as long as the microphone is open, so its real-time factor is a duty cycle. A benchmark races through a file; an audio callback does not. Pacing to the 6.67 ms deadline costs 4.2x per frame, saves 49% of the energy, and leaves the cheapest core placement missing 96% of its deadlines. All SIMD tiers within an architecture family emit byte-identical output. The runtime is released as a dependency-free library.
Gyeong-hee Kim· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
