Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers

• URL: http://arxiv.org/abs/2607.23811v1
• Date: Sun, 26 Jul 2026 19:20:01 GMT
• Title: Memory Efficient Audio Synthesis with Decoupled Temporal Depth Diffusion Transformers
• Abstract summary: Siri Expressive Voices synthesizes rich, conversational speech in real time and entirely on device.<n>A detokenizer converts semantic audio tokens emitted by the foundation model into high-fidelity audio within the tight compute and memory budget of the Apple Matrix Coprocessor (AMX)<n>A single reusable depth decoder with Diffusion Transformer (DiT)-style stage conditioning generates all RVQ levels autoregressively, replacing the dedicated per-level decoders of prior multi-decoder architectures.
• Score: 11.102698205402675
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Siri Expressive Voices synthesize rich, configurable speech in real time and entirely on device, powered by AFM 3 Core Advanced, Apple's most powerful on-device foundation model. This work presents the memory-efficient audio synthesis architecture behind that capability: a detokenizer that converts the semantic audio tokens emitted by the foundation model into high-fidelity audio within the tight compute and memory budget of the Apple Matrix Coprocessor (AMX). We convert semantic audio tokens to a residual vector quantization (RVQ) representation with a three-component design, a streaming encoder, a temporal decoder, and a depth decoder, that systematically decouples temporal and depth processing. A single reusable depth decoder with Diffusion Transformer (DiT)-style stage conditioning generates all RVQ levels autoregressively, replacing the dedicated per-level decoders of prior multi-decoder architectures, while causal sliding window attention with fixed-window key-value caching yields constant memory complexity independent of sequence length. Deployed on the AMX, the detokenizer sustains roughly 10 ms per generation step, about 16x faster than real time, with a peak runtime memory of only 21 MB and 329 MB of on-device assets, enabling continuous streaming synthesis of 20-320 seconds of audio. This constant, small footprint replaces the linear and quadratic memory scaling of conventional transformer- and GAN-based approaches. Ablation studies validate the key architectural components, and audio quality assessment confirms that the architecture maintains synthesis fidelity while achieving efficiency gains over existing methods. Operating at a 1-billion-parameter activation size within AFM 3 Core Advanced, it improves Mean Opinion Score by +0.28 overall (4.15 vs. 3.87) and by +0.42 on conversational speech (4.24 vs. 3.82) over the prior on-device text-to-speech system.

Related papers

• TontaubeV1: Streaming Text-to-Speech with Hierarchical Codec Modeling and Bounded Context [0.04583541422554719]
  TontaubeV1 is a model that preserves natural prosody while enabling streaming from a single consumer GPU.<n>Our design assumes that prosodic structure is largely established when the semantic stream is generated.<n>The model accepts up to one minute of reference audio for voice conditioning and is designed primarily for English and German.
  arXiv  Detail & Related papers  (2026-09-08T13:03:46Z)
• Omni2LoRA: Coherence-Preserving Parametric Memory for Efficient Omni Language Models [65.49950267695267]
  We introduce Omni2LoRA, a framework for efficient parametric memory compression via coherence-preserving context distillation.<n>By converting multimodal memory into a fixed-budget, reusable parameter state, our method drives answer-time multimodal-token load to zero.
  arXiv  Detail & Related papers  (2026-08-10T07:49:10Z)
• TRADE: Transducer-Augmented Decoder for Speech LLM [52.69726809996728]
  Speech Large Language Models (Speech LLMs) lack a principled mechanism for streaming inference.<n>We propose TRADE TRansducer-Augmented DEcoder, which augments a multimodal LLM with a transducer branch that shares the audio encoder.<n>Three design choices make the system accurate, streamable, and long-form capable.
  arXiv  Detail & Related papers  (2026-06-07T07:15:34Z)
• OmniEncoder: See, Hear, and Feel Continuous Motion Like Humans With One Encoder [17.494692126897434]
  We present textbf Omni-Encoder, a unified Transformer backbone designed to co-embed visual and audio signals at a symmetrical 25 fps within a shared latent space.<n> Experiments demonstrate that, compared to the modality-specific baseline Qwen2.5- Omni, Omni-Encoder delivers substantial gains on visual continuous understanding tasks.<n>These results suggest that unified omnivorous encoding offers a promising direction for building omni-modal models.
  arXiv  Detail & Related papers  (2026-05-02T15:55:03Z)
• Pushing the Limits of On-Device Streaming ASR: A Compact, High-Accuracy English Model for Low-Latency Inference [5.685908474026849]
  High-quality automatic speech recognition (ASR) on edge devices requires models that jointly optimize accuracy, latency, and memory footprint while operating entirely on CPU without GPU acceleration.<n>We conduct a systematic empirical study of state-of-the-art ASR architectures, encompassing encoder-decoder, transducer, and LLM-based paradigms, evaluated across batch, chunked, and streaming inference modes.<n>Our recommended configuration, the int4 k-quant variant, achieves 8.20% average streaming WER across eight standard benchmarks, running comfortably faster than real-time on CPU with 0.56 s algorithmic latency.
  arXiv  Detail & Related papers  (2026-04-16T00:04:32Z)
• HH-Codec: High Compression High-fidelity Discrete Neural Codec for Spoken Language Modeling [6.313337261965531]
  We introduce HH-Codec, a neural codecs that achieves extreme compression at 24 tokens per second for 24 kHz audio.<n>Our approach involves a carefully designed Vector Quantization space for Spoken Language Modeling, optimizing compression efficiency while minimizing information loss.<n> HH-Codec achieves state-of-the-art performance in speech reconstruction with an ultra-low bandwidth of 0.3 kbps.
  arXiv  Detail & Related papers  (2025-07-25T02:44:30Z)
• Real-Time Target Sound Extraction [13.526450617545537]
  We present the first neural network model to achieve real-time and streaming target sound extraction. We propose Waveformer, an encoder-decoder architecture with a stack of dilated causal convolution layers as the encoder, and a transformer decoder layer as the decoder.
  arXiv  Detail & Related papers  (2022-11-04T03:51:23Z)
• MAST: Multiscale Audio Spectrogram Transformers [53.06337011259031]
  We present Multiscale Audio Spectrogram Transformer (MAST) for audio classification, which brings the concept of multiscale feature hierarchies to the Audio Spectrogram Transformer (AST) In practice, MAST significantly outperforms AST by an average accuracy of 3.4% across 8 speech and non-speech tasks from the LAPE Benchmark.
  arXiv  Detail & Related papers  (2022-11-02T23:34:12Z)
• High Fidelity Neural Audio Compression [92.4812002532009]
  We introduce a state-of-the-art real-time, high-fidelity, audio leveraging neural networks. It consists in a streaming encoder-decoder architecture with quantized latent space trained in an end-to-end fashion. We simplify and speed-up the training by using a single multiscale spectrogram adversary.
  arXiv  Detail & Related papers  (2022-10-24T17:52:02Z)
• FastLTS: Non-Autoregressive End-to-End Unconstrained Lip-to-Speech Synthesis [77.06890315052563]
  We propose FastLTS, a non-autoregressive end-to-end model which can directly synthesize high-quality speech audios from unconstrained talking videos with low latency. Experiments show that our model achieves $19.76times$ speedup for audio generation compared with the current autoregressive model on input sequences of 3 seconds.
  arXiv  Detail & Related papers  (2022-07-08T10:10:39Z)
• Advanced Long-context End-to-end Speech Recognition Using Context-expanded Transformers [56.56220390953412]
  We extend our prior work by introducing the Conformer architecture to further improve the accuracy. We demonstrate that the extended Transformer provides state-of-the-art end-to-end ASR performance.
  arXiv  Detail & Related papers  (2021-04-19T16:18:00Z)
• Streaming automatic speech recognition with the transformer model [59.58318952000571]
  We propose a transformer based end-to-end ASR system for streaming ASR. We apply time-restricted self-attention for the encoder and triggered attention for the encoder-decoder attention mechanism. Our proposed streaming transformer architecture achieves 2.8% and 7.2% WER for the "clean" and "other" test data of LibriSpeech.
  arXiv  Detail & Related papers  (2020-01-08T18:58:02Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.