# Apple — Compressing Streaming Neural Audio Encoders via Latent-Space Distillation

- Company: Apple (apple.com)
- Announced: 2026-09-24
- Category: research-paper
- Coverage: not counted
- Announcement: yes
- Group: announcements
- Source: https://machinelearning.apple.com/research/latent-space-distillation
- Record: https://forck.live/items/13718-compressing-streaming-neural-audio-encoders-via-latent-space-distillation
- Subject: Machine Learning Research

Apple researchers present a method for compressing streaming neural audio encoders used in on-device dictation through latent-space distillation. The technique trains a student encoder to regress the teacher's pre-quantizer latent representation, achieving 2.8× compression while maintaining within 1.9% relative word error rate of the teacher on most model pairs without fine-tuning.

## Evidence

Verbatim from https://machinelearning.apple.com/research/latent-space-distillation:

> At 2.8× compression the distilled student stays within 1.9% relative WER of its teacher on five of six teacher–student pairs without any fine-tuning, and improves on an independently trained tokenizer of identical capacity by 3.9% relative.

---

Record: https://forck.live/items/13718-compressing-streaming-neural-audio-encoders-via-latent-space-distillation
Catalogue: https://forck.live/llms.txt
Current issue: https://forck.live/feed.md