---
title: "berlin Isaiah (@berlinisaiah) | Hashnode"
description: "berlin Isaiah (@berlinisaiah) writes on Hashnode. Read their latest articles and follow their work."
url: "https://hashnode.com/@berlinisaiah"
---

@berlinisaiah

# berlin Isaiah

@berlinisaiah · Joined May 2026

### [Why Your Flow Matching TTS Model Won't Converge — and the Latent Statistics Behind It](https://voxflash.hashnode.dev/why-your-flow-matching-tts-model-won-t-converge-and-the-latent-statistics-behind-it?utm_source=hashnode&utm_medium=feed)

A statistical deep dive into VAE latent scale, SNR mismatch, and per-channel normalization, with VoxFlash-TTS as a running case study

17 min read · [voxflash.hashnode.dev](https://voxflash.hashnode.dev/why-your-flow-matching-tts-model-won-t-converge-and-the-latent-statistics-behind-it?utm_source=hashnode&utm_medium=feed)

[Jun 30](https://hashnode.com/posts/why-your-flow-matching-tts-model-won-t-converge-and-the-latent-statistics-behind-it/6a435178e40be9c46759e0f2)

### [Why Reference Audio Gets Contaminated by Noise in Masked Diffusion TTS](https://voxflash.hashnode.dev/why-reference-audio-gets-contaminated-by-noise-in-masked-diffusion-tts?utm_source=hashnode&utm_medium=feed)

Why Reference Audio Gets Contaminated by Noise in Masked Diffusion TTS The structural problem with bidirectional self-attention in voice cloning

8 min read · [voxflash.hashnode.dev](https://voxflash.hashnode.dev/why-reference-audio-gets-contaminated-by-noise-in-masked-diffusion-tts?utm_source=hashnode&utm_medium=feed)

[Jun 20](https://hashnode.com/posts/why-reference-audio-gets-contaminated-by-noise-in-masked-diffusion-tts/6a366b6e79bb6ff87690fae7)

### [Why Cross-Attention Breaks in Diffusion-Based TTS — And Three Ways to Fix It](https://voxflash.hashnode.dev/why-cross-attention-breaks-in-diffusion-based-tts-and-three-ways-to-fix-it?utm_source=hashnode&utm_medium=feed)

A deep dive into phoneme alignment, the early-timestep failure problem, and LARoPE

9 min read · [voxflash.hashnode.dev](https://voxflash.hashnode.dev/why-cross-attention-breaks-in-diffusion-based-tts-and-three-ways-to-fix-it?utm_source=hashnode&utm_medium=feed)

[Jun 20](https://hashnode.com/posts/why-cross-attention-breaks-in-diffusion-based-tts-and-three-ways-to-fix-it/6a3667d0b6311ce07a6a371e)

### [Phoneme Alignment in TTS: From Explicit Duration Modeling to Implicit Attention Learning](https://voxflash.hashnode.dev/phoneme-alignment-in-tts-from-explicit-duration-modeling-to-implicit-attention-learning?utm_source=hashnode&utm_medium=feed)

A technical deep dive into how text-to-speech systems solve the phoneme-to-frame alignment problem

32 min read · [voxflash.hashnode.dev](https://voxflash.hashnode.dev/phoneme-alignment-in-tts-from-explicit-duration-modeling-to-implicit-attention-learning?utm_source=hashnode&utm_medium=feed)

[Jun 20](https://hashnode.com/posts/phoneme-alignment-in-tts-from-explicit-duration-modeling-to-implicit-attention-learning/6a365d0a0d3b96a48aea337e)

### [How We Cut TTS Inference to Milliseconds: A Deep Dive into VoxFlash-TTS](https://voxflash.hashnode.dev/how-we-cut-tts-inference-to-milliseconds-a-deep-dive-into-voxflash-tts?utm_source=hashnode&utm_medium=feed)

How VoxFlash-TTS achieves fastest voice cloning via 9 Hz latent diffusion. Zero-shot ZH & EN, consumer GPU, one-command Docker deploy.

5 min read · [voxflash.hashnode.dev](https://voxflash.hashnode.dev/how-we-cut-tts-inference-to-milliseconds-a-deep-dive-into-voxflash-tts?utm_source=hashnode&utm_medium=feed)

[May 25](https://hashnode.com/posts/how-we-cut-tts-inference-to-milliseconds-a-deep-dive-into-voxflash-tts/6a1443ab7d85e6a1afd4ce16)
